Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* A Bison parser, made by GNU Bison 3.0.4. */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* Bison implementation for Yacc-like parsers in C
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
Copyright (C) 1984, 1989-1990, 2000-2015 Free Software Foundation, Inc.
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
This program is free software: you can redistribute it and/or modify
|
|
|
|
it under the terms of the GNU General Public License as published by
|
|
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
|
|
(at your option) any later version.
|
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>. */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* As a special exception, you may create a larger work that contains
|
|
|
|
part or all of the Bison parser skeleton and distribute that work
|
|
|
|
under terms of your choice, so long as that work isn't itself a
|
|
|
|
parser generator using the skeleton or a modified version thereof
|
|
|
|
as a parser skeleton. Alternatively, if you modify or redistribute
|
|
|
|
the parser skeleton itself, you may (at your option) remove this
|
|
|
|
special exception, which will cause the skeleton and the resulting
|
|
|
|
Bison output files to be licensed under the GNU General Public
|
|
|
|
License without this special exception.
|
|
|
|
|
|
|
|
This special exception was added by the Free Software Foundation in
|
|
|
|
version 2.2 of Bison. */
|
|
|
|
|
|
|
|
/* C LALR(1) parser skeleton written by Richard Stallman, by
|
|
|
|
simplifying the original so-called "semantic" parser. */
|
|
|
|
|
|
|
|
/* All symbols defined below should begin with yy or YY, to avoid
|
|
|
|
infringing on user name space. This should be done even for local
|
|
|
|
variables, as they might otherwise be expanded by user macros.
|
|
|
|
There are some unavoidable exceptions within include files to
|
|
|
|
define necessary library symbols; they are noted "INFRINGES ON
|
|
|
|
USER NAME SPACE" below. */
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Identify Bison output. */
|
|
|
|
#define YYBISON 1
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Bison version. */
|
|
|
|
#define YYBISON_VERSION "3.0.4"
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* Skeleton name. */
|
|
|
|
#define YYSKELETON_NAME "yacc.c"
|
|
|
|
|
|
|
|
/* Pure parsers. */
|
|
|
|
#define YYPURE 0
|
|
|
|
|
|
|
|
/* Push parsers. */
|
|
|
|
#define YYPUSH 0
|
|
|
|
|
|
|
|
/* Pull parsers. */
|
|
|
|
#define YYPULL 1
|
|
|
|
|
|
|
|
|
|
|
|
/* Substitute the variable and function names. */
|
|
|
|
#define yyparse ncgparse
|
|
|
|
#define yylex ncglex
|
|
|
|
#define yyerror ncgerror
|
|
|
|
#define yydebug ncgdebug
|
|
|
|
#define yynerrs ncgnerrs
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
#define yylval ncglval
|
|
|
|
#define yychar ncgchar
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Copy the first part of user declarations. */
|
|
|
|
#line 9 "ncgen.y" /* yacc.c:339 */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
#ifdef sccs
|
|
|
|
static char SccsId[] = "$Id: ncgen.y,v 1.34 2010/03/31 18:18:41 dmh Exp $";
|
|
|
|
#endif
|
|
|
|
#include "config.h"
|
|
|
|
#include <string.h>
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#include "netcdf.h"
|
|
|
|
#include "generic.h"
|
|
|
|
#include "ncgen.h"
|
|
|
|
#include "genlib.h" /* for grow_darray() et al */
|
|
|
|
|
|
|
|
typedef struct Symbol { /* symbol table entry */
|
|
|
|
char *name;
|
|
|
|
struct Symbol *next;
|
|
|
|
unsigned is_dim : 1; /* appears as netCDF dimension */
|
|
|
|
unsigned is_var : 1; /* appears as netCDF variable */
|
|
|
|
unsigned is_att : 1; /* appears as netCDF attribute */
|
|
|
|
int dnum; /* handle as a dimension */
|
|
|
|
int vnum; /* handle as a variable */
|
|
|
|
} *YYSTYPE1;
|
|
|
|
|
|
|
|
/* True if string a equals string b*/
|
|
|
|
#ifndef STREQ
|
|
|
|
#define STREQ(a, b) (*(a) == *(b) && strcmp((a), (b)) == 0)
|
|
|
|
#endif
|
|
|
|
#define NC_UNSPECIFIED ((nc_type)0) /* unspecified (as yet) type */
|
|
|
|
|
|
|
|
#define YYSTYPE YYSTYPE1
|
|
|
|
YYSTYPE symlist; /* symbol table: linked list */
|
|
|
|
|
|
|
|
extern int derror_count; /* counts errors in netcdf definition */
|
|
|
|
extern int lineno; /* line number for error messages */
|
|
|
|
|
|
|
|
static int not_a_string; /* whether last constant read was a string */
|
|
|
|
static char termstring[MAXTRST]; /* last terminal string read */
|
|
|
|
static double double_val; /* last double value read */
|
|
|
|
static float float_val; /* last float value read */
|
|
|
|
static int int_val; /* last int value read */
|
|
|
|
static short short_val; /* last short value read */
|
|
|
|
static char char_val; /* last char value read */
|
|
|
|
static signed char byte_val; /* last byte value read */
|
|
|
|
|
|
|
|
static nc_type type_code; /* holds declared type for variables */
|
|
|
|
static nc_type atype_code; /* holds derived type for attributes */
|
|
|
|
static char *netcdfname; /* to construct netcdf file name */
|
|
|
|
static void *att_space; /* pointer to block for attribute values */
|
|
|
|
static nc_type valtype; /* type code for list of attribute values */
|
|
|
|
|
|
|
|
static char *char_valp; /* pointers used to accumulate data values */
|
|
|
|
static signed char *byte_valp;
|
|
|
|
static short *short_valp;
|
|
|
|
static int *int_valp;
|
|
|
|
static float *float_valp;
|
|
|
|
static double *double_valp;
|
|
|
|
static void *rec_cur; /* pointer to where next data value goes */
|
|
|
|
static void *rec_start; /* start of space for data */
|
|
|
|
|
|
|
|
/* Forward declarations */
|
2022-03-02 12:21:24 +08:00
|
|
|
void defatt(void);
|
|
|
|
void equalatt(void);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
#ifdef YYLEX_PARAM
|
|
|
|
int yylex(YYLEX_PARAM);
|
|
|
|
#else
|
2022-03-02 12:21:24 +08:00
|
|
|
int yylex(void);
|
2020-07-30 07:59:19 +08:00
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef vms
|
|
|
|
void yyerror(char*);
|
|
|
|
#else
|
|
|
|
int yyerror(char*);
|
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 150 "ncgeny.c" /* yacc.c:339 */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
# ifndef YY_NULLPTR
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# if defined __cplusplus && 201103L <= __cplusplus
|
|
|
|
# define YY_NULLPTR nullptr
|
2020-07-30 07:59:19 +08:00
|
|
|
# else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YY_NULLPTR 0
|
2020-07-30 07:59:19 +08:00
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Enabling verbose error messages. */
|
|
|
|
#ifdef YYERROR_VERBOSE
|
|
|
|
# undef YYERROR_VERBOSE
|
|
|
|
# define YYERROR_VERBOSE 1
|
|
|
|
#else
|
|
|
|
# define YYERROR_VERBOSE 0
|
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* In a future release of Bison, this section will be replaced
|
|
|
|
by #include "ncgeny.h". */
|
|
|
|
#ifndef YY_NCG_NCGEN_TAB_H_INCLUDED
|
|
|
|
# define YY_NCG_NCGEN_TAB_H_INCLUDED
|
|
|
|
/* Debug traces. */
|
|
|
|
#ifndef YYDEBUG
|
|
|
|
# define YYDEBUG 1
|
|
|
|
#endif
|
|
|
|
#if YYDEBUG
|
|
|
|
extern int ncgdebug;
|
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Token type. */
|
|
|
|
#ifndef YYTOKENTYPE
|
|
|
|
# define YYTOKENTYPE
|
|
|
|
enum yytokentype
|
|
|
|
{
|
|
|
|
NC_UNLIMITED_K = 258,
|
|
|
|
BYTE_K = 259,
|
|
|
|
CHAR_K = 260,
|
|
|
|
SHORT_K = 261,
|
|
|
|
INT_K = 262,
|
|
|
|
FLOAT_K = 263,
|
|
|
|
DOUBLE_K = 264,
|
|
|
|
IDENT = 265,
|
|
|
|
TERMSTRING = 266,
|
|
|
|
BYTE_CONST = 267,
|
|
|
|
CHAR_CONST = 268,
|
|
|
|
SHORT_CONST = 269,
|
|
|
|
INT_CONST = 270,
|
|
|
|
FLOAT_CONST = 271,
|
|
|
|
DOUBLE_CONST = 272,
|
|
|
|
DIMENSIONS = 273,
|
|
|
|
VARIABLES = 274,
|
|
|
|
NETCDF = 275,
|
|
|
|
DATA = 276,
|
|
|
|
FILLVALUE = 277
|
|
|
|
};
|
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Value type. */
|
|
|
|
#if ! defined YYSTYPE && ! defined YYSTYPE_IS_DECLARED
|
|
|
|
typedef int YYSTYPE;
|
|
|
|
# define YYSTYPE_IS_TRIVIAL 1
|
|
|
|
# define YYSTYPE_IS_DECLARED 1
|
2021-09-03 07:04:26 +08:00
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
extern YYSTYPE ncglval;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
int ncgparse (void);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#endif /* !YY_NCG_NCGEN_TAB_H_INCLUDED */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Copy the second part of user declarations. */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 224 "ncgeny.c" /* yacc.c:358 */
|
|
|
|
|
|
|
|
#ifdef short
|
|
|
|
# undef short
|
2020-07-30 07:59:19 +08:00
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#ifdef YYTYPE_UINT8
|
|
|
|
typedef YYTYPE_UINT8 yytype_uint8;
|
|
|
|
#else
|
2021-09-03 07:04:26 +08:00
|
|
|
typedef unsigned char yytype_uint8;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef YYTYPE_INT8
|
|
|
|
typedef YYTYPE_INT8 yytype_int8;
|
2020-07-30 07:59:19 +08:00
|
|
|
#else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
typedef signed char yytype_int8;
|
2020-07-30 07:59:19 +08:00
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#ifdef YYTYPE_UINT16
|
|
|
|
typedef YYTYPE_UINT16 yytype_uint16;
|
2020-07-30 07:59:19 +08:00
|
|
|
#else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
typedef unsigned short int yytype_uint16;
|
2021-09-03 07:04:26 +08:00
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#ifdef YYTYPE_INT16
|
|
|
|
typedef YYTYPE_INT16 yytype_int16;
|
|
|
|
#else
|
|
|
|
typedef short int yytype_int16;
|
2020-07-30 07:59:19 +08:00
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifndef YYSIZE_T
|
|
|
|
# ifdef __SIZE_TYPE__
|
|
|
|
# define YYSIZE_T __SIZE_TYPE__
|
|
|
|
# elif defined size_t
|
|
|
|
# define YYSIZE_T size_t
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# elif ! defined YYSIZE_T
|
2020-07-30 07:59:19 +08:00
|
|
|
# include <stddef.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# define YYSIZE_T size_t
|
|
|
|
# else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YYSIZE_T unsigned int
|
2020-07-30 07:59:19 +08:00
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define YYSIZE_MAXIMUM ((YYSIZE_T) -1)
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
#ifndef YY_
|
|
|
|
# if defined YYENABLE_NLS && YYENABLE_NLS
|
|
|
|
# if ENABLE_NLS
|
|
|
|
# include <libintl.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# define YY_(Msgid) dgettext ("bison-runtime", Msgid)
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# ifndef YY_
|
|
|
|
# define YY_(Msgid) Msgid
|
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#ifndef YY_ATTRIBUTE
|
|
|
|
# if (defined __GNUC__ \
|
|
|
|
&& (2 < __GNUC__ || (__GNUC__ == 2 && 96 <= __GNUC_MINOR__))) \
|
|
|
|
|| defined __SUNPRO_C && 0x5110 <= __SUNPRO_C
|
|
|
|
# define YY_ATTRIBUTE(Spec) __attribute__(Spec)
|
2021-09-03 07:04:26 +08:00
|
|
|
# else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YY_ATTRIBUTE(Spec) /* empty */
|
2021-09-03 07:04:26 +08:00
|
|
|
# endif
|
2020-07-30 07:59:19 +08:00
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#ifndef YY_ATTRIBUTE_PURE
|
|
|
|
# define YY_ATTRIBUTE_PURE YY_ATTRIBUTE ((__pure__))
|
|
|
|
#endif
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
#ifndef YY_ATTRIBUTE_UNUSED
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YY_ATTRIBUTE_UNUSED YY_ATTRIBUTE ((__unused__))
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#if !defined _Noreturn \
|
|
|
|
&& (!defined __STDC_VERSION__ || __STDC_VERSION__ < 201112)
|
|
|
|
# if defined _MSC_VER && 1200 <= _MSC_VER
|
|
|
|
# define _Noreturn __declspec (noreturn)
|
2020-07-30 07:59:19 +08:00
|
|
|
# else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define _Noreturn YY_ATTRIBUTE ((__noreturn__))
|
2020-07-30 07:59:19 +08:00
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* Suppress unused-variable warnings by "using" E. */
|
|
|
|
#if ! defined lint || defined __GNUC__
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YYUSE(E) ((void) (E))
|
2020-07-30 07:59:19 +08:00
|
|
|
#else
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YYUSE(E) /* empty */
|
2020-07-30 07:59:19 +08:00
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if defined __GNUC__ && 407 <= __GNUC__ * 100 + __GNUC_MINOR__
|
2020-07-30 07:59:19 +08:00
|
|
|
/* Suppress an incorrect diagnostic about yylval being uninitialized. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YY_IGNORE_MAYBE_UNINITIALIZED_BEGIN \
|
|
|
|
_Pragma ("GCC diagnostic push") \
|
|
|
|
_Pragma ("GCC diagnostic ignored \"-Wuninitialized\"")\
|
2020-07-30 07:59:19 +08:00
|
|
|
_Pragma ("GCC diagnostic ignored \"-Wmaybe-uninitialized\"")
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YY_IGNORE_MAYBE_UNINITIALIZED_END \
|
2020-07-30 07:59:19 +08:00
|
|
|
_Pragma ("GCC diagnostic pop")
|
|
|
|
#else
|
|
|
|
# define YY_INITIAL_VALUE(Value) Value
|
|
|
|
#endif
|
|
|
|
#ifndef YY_IGNORE_MAYBE_UNINITIALIZED_BEGIN
|
|
|
|
# define YY_IGNORE_MAYBE_UNINITIALIZED_BEGIN
|
|
|
|
# define YY_IGNORE_MAYBE_UNINITIALIZED_END
|
|
|
|
#endif
|
|
|
|
#ifndef YY_INITIAL_VALUE
|
|
|
|
# define YY_INITIAL_VALUE(Value) /* Nothing. */
|
|
|
|
#endif
|
|
|
|
|
2021-09-03 07:04:26 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if ! defined yyoverflow || YYERROR_VERBOSE
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* The parser invokes alloca or malloc; define the necessary symbols. */
|
|
|
|
|
|
|
|
# ifdef YYSTACK_USE_ALLOCA
|
|
|
|
# if YYSTACK_USE_ALLOCA
|
|
|
|
# ifdef __GNUC__
|
|
|
|
# define YYSTACK_ALLOC __builtin_alloca
|
|
|
|
# elif defined __BUILTIN_VA_ARG_INCR
|
|
|
|
# include <alloca.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# elif defined _AIX
|
|
|
|
# define YYSTACK_ALLOC __alloca
|
|
|
|
# elif defined _MSC_VER
|
|
|
|
# include <malloc.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# define alloca _alloca
|
|
|
|
# else
|
|
|
|
# define YYSTACK_ALLOC alloca
|
|
|
|
# if ! defined _ALLOCA_H && ! defined EXIT_SUCCESS
|
|
|
|
# include <stdlib.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
/* Use EXIT_SUCCESS as a witness for stdlib.h. */
|
|
|
|
# ifndef EXIT_SUCCESS
|
|
|
|
# define EXIT_SUCCESS 0
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
|
|
|
|
# ifdef YYSTACK_ALLOC
|
|
|
|
/* Pacify GCC's 'empty if-body' warning. */
|
|
|
|
# define YYSTACK_FREE(Ptr) do { /* empty */; } while (0)
|
|
|
|
# ifndef YYSTACK_ALLOC_MAXIMUM
|
|
|
|
/* The OS might guarantee only one guard page at the bottom of the stack,
|
|
|
|
and a page size can be as small as 4096 bytes. So we cannot safely
|
|
|
|
invoke alloca (N) if N exceeds 4096. Use a slightly smaller number
|
|
|
|
to allow for a few compiler-allocated temporary stack slots. */
|
|
|
|
# define YYSTACK_ALLOC_MAXIMUM 4032 /* reasonable circa 2006 */
|
|
|
|
# endif
|
|
|
|
# else
|
|
|
|
# define YYSTACK_ALLOC YYMALLOC
|
|
|
|
# define YYSTACK_FREE YYFREE
|
|
|
|
# ifndef YYSTACK_ALLOC_MAXIMUM
|
|
|
|
# define YYSTACK_ALLOC_MAXIMUM YYSIZE_MAXIMUM
|
|
|
|
# endif
|
|
|
|
# if (defined __cplusplus && ! defined EXIT_SUCCESS \
|
|
|
|
&& ! ((defined YYMALLOC || defined malloc) \
|
|
|
|
&& (defined YYFREE || defined free)))
|
|
|
|
# include <stdlib.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# ifndef EXIT_SUCCESS
|
|
|
|
# define EXIT_SUCCESS 0
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# ifndef YYMALLOC
|
|
|
|
# define YYMALLOC malloc
|
|
|
|
# if ! defined malloc && ! defined EXIT_SUCCESS
|
|
|
|
void *malloc (YYSIZE_T); /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# ifndef YYFREE
|
|
|
|
# define YYFREE free
|
|
|
|
# if ! defined free && ! defined EXIT_SUCCESS
|
|
|
|
void free (void *); /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
# endif
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#endif /* ! defined yyoverflow || YYERROR_VERBOSE */
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
#if (! defined yyoverflow \
|
|
|
|
&& (! defined __cplusplus \
|
|
|
|
|| (defined YYSTYPE_IS_TRIVIAL && YYSTYPE_IS_TRIVIAL)))
|
|
|
|
|
|
|
|
/* A type that is properly aligned for any stack member. */
|
|
|
|
union yyalloc
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yytype_int16 yyss_alloc;
|
2020-07-30 07:59:19 +08:00
|
|
|
YYSTYPE yyvs_alloc;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* The size of the maximum gap between one aligned stack and the next. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YYSTACK_GAP_MAXIMUM (sizeof (union yyalloc) - 1)
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* The size of an array large to enough to hold all stacks, each with
|
|
|
|
N elements. */
|
|
|
|
# define YYSTACK_BYTES(N) \
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
((N) * (sizeof (yytype_int16) + sizeof (YYSTYPE)) \
|
2020-07-30 07:59:19 +08:00
|
|
|
+ YYSTACK_GAP_MAXIMUM)
|
|
|
|
|
|
|
|
# define YYCOPY_NEEDED 1
|
|
|
|
|
|
|
|
/* Relocate STACK from its old location to the new one. The
|
|
|
|
local variables YYSIZE and YYSTACKSIZE give the old and new number of
|
|
|
|
elements in the stack, and YYPTR gives the new location of the
|
|
|
|
stack. Advance YYPTR to a properly aligned location for the next
|
|
|
|
stack. */
|
|
|
|
# define YYSTACK_RELOCATE(Stack_alloc, Stack) \
|
|
|
|
do \
|
|
|
|
{ \
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYSIZE_T yynewbytes; \
|
2020-07-30 07:59:19 +08:00
|
|
|
YYCOPY (&yyptr->Stack_alloc, Stack, yysize); \
|
|
|
|
Stack = &yyptr->Stack_alloc; \
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yynewbytes = yystacksize * sizeof (*Stack) + YYSTACK_GAP_MAXIMUM; \
|
|
|
|
yyptr += yynewbytes / sizeof (*yyptr); \
|
2020-07-30 07:59:19 +08:00
|
|
|
} \
|
|
|
|
while (0)
|
|
|
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#if defined YYCOPY_NEEDED && YYCOPY_NEEDED
|
|
|
|
/* Copy COUNT objects from SRC to DST. The source and destination do
|
|
|
|
not overlap. */
|
|
|
|
# ifndef YYCOPY
|
|
|
|
# if defined __GNUC__ && 1 < __GNUC__
|
|
|
|
# define YYCOPY(Dst, Src, Count) \
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
__builtin_memcpy (Dst, Src, (Count) * sizeof (*(Src)))
|
2020-07-30 07:59:19 +08:00
|
|
|
# else
|
|
|
|
# define YYCOPY(Dst, Src, Count) \
|
|
|
|
do \
|
|
|
|
{ \
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYSIZE_T yyi; \
|
2020-07-30 07:59:19 +08:00
|
|
|
for (yyi = 0; yyi < (Count); yyi++) \
|
|
|
|
(Dst)[yyi] = (Src)[yyi]; \
|
|
|
|
} \
|
|
|
|
while (0)
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
#endif /* !YYCOPY_NEEDED */
|
|
|
|
|
|
|
|
/* YYFINAL -- State number of the termination state. */
|
|
|
|
#define YYFINAL 4
|
|
|
|
/* YYLAST -- Last index in YYTABLE. */
|
|
|
|
#define YYLAST 78
|
|
|
|
|
|
|
|
/* YYNTOKENS -- Number of terminals. */
|
|
|
|
#define YYNTOKENS 31
|
|
|
|
/* YYNNTS -- Number of nonterminals. */
|
|
|
|
#define YYNNTS 42
|
|
|
|
/* YYNRULES -- Number of rules. */
|
|
|
|
#define YYNRULES 79
|
|
|
|
/* YYNSTATES -- Number of states. */
|
|
|
|
#define YYNSTATES 112
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYTRANSLATE[YYX] -- Symbol number corresponding to YYX as returned
|
|
|
|
by yylex, with out-of-bounds checking. */
|
|
|
|
#define YYUNDEFTOK 2
|
2020-07-30 07:59:19 +08:00
|
|
|
#define YYMAXUTOK 277
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define YYTRANSLATE(YYX) \
|
|
|
|
((unsigned int) (YYX) <= YYMAXUTOK ? yytranslate[YYX] : YYUNDEFTOK)
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* YYTRANSLATE[TOKEN-NUM] -- Symbol number corresponding to TOKEN-NUM
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
as returned by yylex, without out-of-bounds checking. */
|
|
|
|
static const yytype_uint8 yytranslate[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
0, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
28, 29, 2, 2, 26, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 30, 25,
|
|
|
|
2, 27, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 23, 2, 24, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
|
2, 2, 2, 2, 2, 2, 1, 2, 3, 4,
|
|
|
|
5, 6, 7, 8, 9, 10, 11, 12, 13, 14,
|
|
|
|
15, 16, 17, 18, 19, 20, 21, 22
|
|
|
|
};
|
|
|
|
|
|
|
|
#if YYDEBUG
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYRLINE[YYN] -- Source line where rule number YYN was defined. */
|
|
|
|
static const yytype_uint16 yyrline[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
2021-09-29 04:03:19 +08:00
|
|
|
0, 117, 117, 120, 115, 133, 134, 136, 137, 139,
|
|
|
|
140, 142, 148, 159, 167, 184, 186, 187, 188, 190,
|
|
|
|
191, 193, 193, 193, 195, 196, 198, 200, 201, 202,
|
|
|
|
203, 204, 205, 207, 208, 211, 210, 249, 251, 252,
|
|
|
|
254, 255, 257, 279, 278, 288, 287, 297, 299, 305,
|
|
|
|
315, 326, 327, 329, 338, 344, 357, 363, 369, 375,
|
|
|
|
381, 389, 390, 391, 394, 395, 398, 397, 464, 465,
|
|
|
|
468, 468, 520, 546, 601, 627, 653, 679, 705, 734
|
2020-07-30 07:59:19 +08:00
|
|
|
};
|
|
|
|
#endif
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if YYDEBUG || YYERROR_VERBOSE || 0
|
2020-07-30 07:59:19 +08:00
|
|
|
/* YYTNAME[SYMBOL-NUM] -- String name of the symbol SYMBOL-NUM.
|
|
|
|
First, the terminals, then, starting at YYNTOKENS, nonterminals. */
|
|
|
|
static const char *const yytname[] =
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
"$end", "error", "$undefined", "NC_UNLIMITED_K", "BYTE_K", "CHAR_K",
|
|
|
|
"SHORT_K", "INT_K", "FLOAT_K", "DOUBLE_K", "IDENT", "TERMSTRING",
|
|
|
|
"BYTE_CONST", "CHAR_CONST", "SHORT_CONST", "INT_CONST", "FLOAT_CONST",
|
|
|
|
"DOUBLE_CONST", "DIMENSIONS", "VARIABLES", "NETCDF", "DATA", "FILLVALUE",
|
|
|
|
"'{'", "'}'", "';'", "','", "'='", "'('", "')'", "':'", "$accept",
|
|
|
|
"ncdesc", "$@1", "$@2", "dimsection", "dimdecls", "dimdecline",
|
|
|
|
"dimdecl", "dimd", "dim", "vasection", "vadecls", "vadecl", "gattdecls",
|
|
|
|
"vardecl", "type", "varlist", "varspec", "$@3", "var", "dimspec",
|
|
|
|
"dimlist", "vdim", "attdecl", "$@4", "gattdecl", "$@5", "att", "gatt",
|
|
|
|
"avar", "attr", "attvallist", "aconst", "attconst", "datasection",
|
|
|
|
"datadecls", "datadecl", "$@6", "constlist", "dconst", "$@7", "const", YY_NULLPTR
|
2020-07-30 07:59:19 +08:00
|
|
|
};
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#endif
|
2021-09-03 07:04:26 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# ifdef YYPRINT
|
|
|
|
/* YYTOKNUM[NUM] -- (External) token number corresponding to the
|
|
|
|
(internal) symbol number NUM (which must be that of a token). */
|
|
|
|
static const yytype_uint16 yytoknum[] =
|
2021-09-03 07:04:26 +08:00
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
0, 256, 257, 258, 259, 260, 261, 262, 263, 264,
|
|
|
|
265, 266, 267, 268, 269, 270, 271, 272, 273, 274,
|
|
|
|
275, 276, 277, 123, 125, 59, 44, 61, 40, 41,
|
|
|
|
58
|
|
|
|
};
|
|
|
|
# endif
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define YYPACT_NINF -73
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define yypact_value_is_default(Yystate) \
|
|
|
|
(!!((Yystate) == (-73)))
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define YYTABLE_NINF -1
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define yytable_value_is_error(Yytable_value) \
|
2020-07-30 07:59:19 +08:00
|
|
|
0
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYPACT[STATE-NUM] -- Index in YYTABLE of the portion describing
|
|
|
|
STATE-NUM. */
|
2020-07-30 07:59:19 +08:00
|
|
|
static const yytype_int8 yypact[] =
|
|
|
|
{
|
|
|
|
10, 3, 31, -73, -73, 19, 36, 6, -73, 36,
|
|
|
|
7, -73, 20, -73, -3, 38, -73, 21, 24, -73,
|
|
|
|
9, -73, 36, 5, -73, -73, -73, -73, -73, -73,
|
|
|
|
-73, -3, 25, -73, 42, -73, -73, -73, -73, 23,
|
|
|
|
-73, -73, 33, 30, -73, 29, -73, -73, -73, -73,
|
|
|
|
-73, 32, -73, 34, -73, -73, 35, 38, 42, 37,
|
|
|
|
-73, 28, -73, 42, 39, 28, -73, -73, 42, 40,
|
|
|
|
-73, -73, -73, -73, -73, -73, -73, -73, 43, -73,
|
|
|
|
-73, -73, 36, -73, 43, 41, 45, -73, 28, -73,
|
|
|
|
-17, -73, -73, -73, -73, 36, -73, 46, -73, 2,
|
|
|
|
-73, -73, -73, -73, -73, -73, -73, -73, -73, -73,
|
|
|
|
-73, -73
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYDEFACT[STATE-NUM] -- Default reduction number in state STATE-NUM.
|
|
|
|
Performed when YYTABLE does not specify something else to do. Zero
|
|
|
|
means the default is an error. */
|
|
|
|
static const yytype_uint8 yydefact[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
0, 0, 0, 2, 1, 5, 0, 16, 15, 6,
|
|
|
|
0, 9, 0, 14, 0, 0, 3, 18, 0, 45,
|
|
|
|
0, 7, 0, 0, 27, 28, 29, 30, 31, 32,
|
|
|
|
37, 17, 0, 21, 0, 49, 22, 23, 43, 0,
|
|
|
|
50, 48, 61, 0, 24, 0, 8, 10, 13, 11,
|
|
|
|
12, 0, 19, 26, 33, 35, 0, 0, 63, 0,
|
|
|
|
25, 0, 20, 0, 38, 0, 47, 66, 62, 0,
|
|
|
|
4, 55, 56, 54, 57, 58, 59, 60, 46, 51,
|
|
|
|
53, 34, 0, 36, 44, 0, 0, 64, 0, 42,
|
|
|
|
0, 40, 70, 65, 52, 0, 39, 67, 68, 0,
|
|
|
|
41, 70, 73, 74, 72, 75, 76, 77, 78, 79,
|
|
|
|
71, 69
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYPGOTO[NTERM-NUM]. */
|
2020-07-30 07:59:19 +08:00
|
|
|
static const yytype_int8 yypgoto[] =
|
|
|
|
{
|
|
|
|
-73, -73, -73, -73, -73, -73, 49, 44, -73, -72,
|
|
|
|
-73, -73, 47, -73, -73, -73, -73, -4, -73, -34,
|
|
|
|
-73, -73, -32, -73, -73, 4, -73, -73, -73, -30,
|
|
|
|
14, -1, -15, -73, -73, -73, 8, -73, -73, -27,
|
|
|
|
-73, -73
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYDEFGOTO[NTERM-NUM]. */
|
2020-07-30 07:59:19 +08:00
|
|
|
static const yytype_int8 yydefgoto[] =
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
-1, 2, 5, 42, 7, 9, 10, 11, 12, 13,
|
2020-07-30 07:59:19 +08:00
|
|
|
16, 31, 32, 17, 33, 34, 53, 54, 64, 35,
|
|
|
|
83, 90, 91, 36, 56, 37, 45, 38, 19, 39,
|
|
|
|
41, 78, 79, 80, 59, 68, 69, 85, 97, 98,
|
|
|
|
99, 110
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYTABLE[YYPACT[STATE-NUM]] -- What to do in state STATE-NUM. If
|
|
|
|
positive, shift that token. If negative, reduce the rule whose
|
|
|
|
number is the opposite. If YYTABLE_NINF, syntax error. */
|
|
|
|
static const yytype_uint8 yytable[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
55, 24, 25, 26, 27, 28, 29, 30, 48, 95,
|
|
|
|
89, 18, 96, 102, 103, 104, 105, 106, 107, 108,
|
|
|
|
49, 43, 50, 89, 109, 14, 3, 15, 67, 55,
|
|
|
|
1, 4, 21, 22, 46, 22, 15, 6, 67, 71,
|
|
|
|
72, 73, 74, 75, 76, 77, 8, 23, 40, 44,
|
|
|
|
52, 15, 30, 57, 58, 60, 61, 62, 20, 81,
|
|
|
|
63, 70, 65, 100, 84, 87, 47, 82, 92, 88,
|
|
|
|
93, 66, 101, 94, 111, 0, 86, 0, 51
|
|
|
|
};
|
|
|
|
|
|
|
|
static const yytype_int8 yycheck[] =
|
|
|
|
{
|
|
|
|
34, 4, 5, 6, 7, 8, 9, 10, 3, 26,
|
|
|
|
82, 7, 29, 11, 12, 13, 14, 15, 16, 17,
|
|
|
|
15, 17, 17, 95, 22, 19, 23, 30, 58, 63,
|
|
|
|
20, 0, 25, 26, 25, 26, 30, 18, 68, 11,
|
|
|
|
12, 13, 14, 15, 16, 17, 10, 27, 10, 25,
|
|
|
|
25, 30, 10, 30, 21, 25, 27, 25, 9, 63,
|
|
|
|
26, 24, 27, 95, 65, 25, 22, 28, 27, 26,
|
|
|
|
25, 57, 26, 88, 101, -1, 68, -1, 31
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYSTOS[STATE-NUM] -- The (internal number of the) accessing
|
|
|
|
symbol of state STATE-NUM. */
|
|
|
|
static const yytype_uint8 yystos[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
0, 20, 32, 23, 0, 33, 18, 35, 10, 36,
|
|
|
|
37, 38, 39, 40, 19, 30, 41, 44, 56, 59,
|
|
|
|
37, 25, 26, 27, 4, 5, 6, 7, 8, 9,
|
|
|
|
10, 42, 43, 45, 46, 50, 54, 56, 58, 60,
|
|
|
|
10, 61, 34, 56, 25, 57, 25, 38, 3, 15,
|
|
|
|
17, 43, 25, 47, 48, 50, 55, 30, 21, 65,
|
|
|
|
25, 27, 25, 26, 49, 27, 61, 60, 66, 67,
|
|
|
|
24, 11, 12, 13, 14, 15, 16, 17, 62, 63,
|
|
|
|
64, 48, 28, 51, 62, 68, 67, 25, 26, 40,
|
|
|
|
52, 53, 27, 25, 63, 26, 29, 69, 70, 71,
|
|
|
|
53, 26, 11, 12, 13, 14, 15, 16, 17, 22,
|
|
|
|
72, 70
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYR1[YYN] -- Symbol number of symbol that rule YYN derives. */
|
|
|
|
static const yytype_uint8 yyr1[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
0, 31, 33, 34, 32, 35, 35, 36, 36, 37,
|
|
|
|
37, 38, 38, 38, 39, 40, 41, 41, 41, 42,
|
|
|
|
42, 43, 43, 43, 44, 44, 45, 46, 46, 46,
|
|
|
|
46, 46, 46, 47, 47, 49, 48, 50, 51, 51,
|
|
|
|
52, 52, 53, 55, 54, 57, 56, 58, 59, 60,
|
|
|
|
61, 62, 62, 63, 64, 64, 64, 64, 64, 64,
|
|
|
|
64, 65, 65, 65, 66, 66, 68, 67, 69, 69,
|
|
|
|
71, 70, 72, 72, 72, 72, 72, 72, 72, 72
|
|
|
|
};
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYR2[YYN] -- Number of symbols on the right hand side of rule YYN. */
|
|
|
|
static const yytype_uint8 yyr2[] =
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
0, 2, 0, 0, 8, 0, 2, 2, 3, 1,
|
|
|
|
3, 3, 3, 3, 1, 1, 0, 2, 1, 2,
|
|
|
|
3, 1, 1, 1, 2, 3, 2, 1, 1, 1,
|
|
|
|
1, 1, 1, 1, 3, 0, 3, 1, 0, 3,
|
|
|
|
1, 3, 1, 0, 4, 0, 4, 3, 2, 1,
|
|
|
|
1, 1, 3, 1, 1, 1, 1, 1, 1, 1,
|
|
|
|
1, 0, 2, 1, 2, 3, 0, 4, 1, 3,
|
|
|
|
0, 2, 1, 1, 1, 1, 1, 1, 1, 1
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
#define yyerrok (yyerrstatus = 0)
|
|
|
|
#define yyclearin (yychar = YYEMPTY)
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define YYEMPTY (-2)
|
|
|
|
#define YYEOF 0
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
#define YYACCEPT goto yyacceptlab
|
|
|
|
#define YYABORT goto yyabortlab
|
|
|
|
#define YYERROR goto yyerrorlab
|
|
|
|
|
|
|
|
|
|
|
|
#define YYRECOVERING() (!!yyerrstatus)
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#define YYBACKUP(Token, Value) \
|
|
|
|
do \
|
|
|
|
if (yychar == YYEMPTY) \
|
|
|
|
{ \
|
|
|
|
yychar = (Token); \
|
|
|
|
yylval = (Value); \
|
|
|
|
YYPOPSTACK (yylen); \
|
|
|
|
yystate = *yyssp; \
|
|
|
|
goto yybackup; \
|
|
|
|
} \
|
|
|
|
else \
|
|
|
|
{ \
|
|
|
|
yyerror (YY_("syntax error: cannot back up")); \
|
|
|
|
YYERROR; \
|
|
|
|
} \
|
|
|
|
while (0)
|
|
|
|
|
|
|
|
/* Error token number */
|
|
|
|
#define YYTERROR 1
|
|
|
|
#define YYERRCODE 256
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
/* Enable debugging if requested. */
|
|
|
|
#if YYDEBUG
|
|
|
|
|
|
|
|
# ifndef YYFPRINTF
|
|
|
|
# include <stdio.h> /* INFRINGES ON USER NAME SPACE */
|
|
|
|
# define YYFPRINTF fprintf
|
|
|
|
# endif
|
|
|
|
|
|
|
|
# define YYDPRINTF(Args) \
|
|
|
|
do { \
|
|
|
|
if (yydebug) \
|
|
|
|
YYFPRINTF Args; \
|
|
|
|
} while (0)
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* This macro is provided for backward compatibility. */
|
|
|
|
#ifndef YY_LOCATION_PRINT
|
|
|
|
# define YY_LOCATION_PRINT(File, Loc) ((void) 0)
|
|
|
|
#endif
|
2021-12-24 13:18:56 +08:00
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YY_SYMBOL_PRINT(Title, Type, Value, Location) \
|
2020-07-30 07:59:19 +08:00
|
|
|
do { \
|
|
|
|
if (yydebug) \
|
|
|
|
{ \
|
|
|
|
YYFPRINTF (stderr, "%s ", Title); \
|
|
|
|
yy_symbol_print (stderr, \
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
Type, Value); \
|
2020-07-30 07:59:19 +08:00
|
|
|
YYFPRINTF (stderr, "\n"); \
|
|
|
|
} \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/*----------------------------------------.
|
|
|
|
| Print this symbol's value on YYOUTPUT. |
|
|
|
|
`----------------------------------------*/
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
static void
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yy_symbol_value_print (FILE *yyoutput, int yytype, YYSTYPE const * const yyvaluep)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
FILE *yyo = yyoutput;
|
|
|
|
YYUSE (yyo);
|
2020-07-30 07:59:19 +08:00
|
|
|
if (!yyvaluep)
|
|
|
|
return;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# ifdef YYPRINT
|
|
|
|
if (yytype < YYNTOKENS)
|
|
|
|
YYPRINT (yyoutput, yytoknum[yytype], *yyvaluep);
|
|
|
|
# endif
|
|
|
|
YYUSE (yytype);
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/*--------------------------------.
|
|
|
|
| Print this symbol on YYOUTPUT. |
|
|
|
|
`--------------------------------*/
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
static void
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yy_symbol_print (FILE *yyoutput, int yytype, YYSTYPE const * const yyvaluep)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYFPRINTF (yyoutput, "%s %s (",
|
|
|
|
yytype < YYNTOKENS ? "token" : "nterm", yytname[yytype]);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yy_symbol_value_print (yyoutput, yytype, yyvaluep);
|
|
|
|
YYFPRINTF (yyoutput, ")");
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*------------------------------------------------------------------.
|
|
|
|
| yy_stack_print -- Print the state stack from its BOTTOM up to its |
|
|
|
|
| TOP (included). |
|
|
|
|
`------------------------------------------------------------------*/
|
|
|
|
|
|
|
|
static void
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yy_stack_print (yytype_int16 *yybottom, yytype_int16 *yytop)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
YYFPRINTF (stderr, "Stack now");
|
|
|
|
for (; yybottom <= yytop; yybottom++)
|
|
|
|
{
|
|
|
|
int yybot = *yybottom;
|
|
|
|
YYFPRINTF (stderr, " %d", yybot);
|
|
|
|
}
|
|
|
|
YYFPRINTF (stderr, "\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
# define YY_STACK_PRINT(Bottom, Top) \
|
|
|
|
do { \
|
|
|
|
if (yydebug) \
|
|
|
|
yy_stack_print ((Bottom), (Top)); \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
|
|
|
|
/*------------------------------------------------.
|
|
|
|
| Report that the YYRULE is going to be reduced. |
|
|
|
|
`------------------------------------------------*/
|
|
|
|
|
|
|
|
static void
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yy_reduce_print (yytype_int16 *yyssp, YYSTYPE *yyvsp, int yyrule)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
unsigned long int yylno = yyrline[yyrule];
|
2020-07-30 07:59:19 +08:00
|
|
|
int yynrhs = yyr2[yyrule];
|
|
|
|
int yyi;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYFPRINTF (stderr, "Reducing stack by rule %d (line %lu):\n",
|
2020-07-30 07:59:19 +08:00
|
|
|
yyrule - 1, yylno);
|
|
|
|
/* The symbols being reduced. */
|
|
|
|
for (yyi = 0; yyi < yynrhs; yyi++)
|
|
|
|
{
|
|
|
|
YYFPRINTF (stderr, " $%d = ", yyi + 1);
|
|
|
|
yy_symbol_print (stderr,
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yystos[yyssp[yyi + 1 - yynrhs]],
|
|
|
|
&(yyvsp[(yyi + 1) - (yynrhs)])
|
|
|
|
);
|
2020-07-30 07:59:19 +08:00
|
|
|
YYFPRINTF (stderr, "\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
# define YY_REDUCE_PRINT(Rule) \
|
|
|
|
do { \
|
|
|
|
if (yydebug) \
|
|
|
|
yy_reduce_print (yyssp, yyvsp, Rule); \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
/* Nonzero means print parse trace. It is left uninitialized so that
|
|
|
|
multiple parsers can coexist. */
|
|
|
|
int yydebug;
|
|
|
|
#else /* !YYDEBUG */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
# define YYDPRINTF(Args)
|
|
|
|
# define YY_SYMBOL_PRINT(Title, Type, Value, Location)
|
2020-07-30 07:59:19 +08:00
|
|
|
# define YY_STACK_PRINT(Bottom, Top)
|
|
|
|
# define YY_REDUCE_PRINT(Rule)
|
|
|
|
#endif /* !YYDEBUG */
|
|
|
|
|
|
|
|
|
|
|
|
/* YYINITDEPTH -- initial size of the parser's stacks. */
|
|
|
|
#ifndef YYINITDEPTH
|
|
|
|
# define YYINITDEPTH 200
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* YYMAXDEPTH -- maximum size the stacks can grow to (effective only
|
|
|
|
if the built-in stack extension method is used).
|
|
|
|
|
|
|
|
Do not make this value too large; the results are undefined if
|
|
|
|
YYSTACK_ALLOC_MAXIMUM < YYSTACK_BYTES (YYMAXDEPTH)
|
|
|
|
evaluated with infinite-precision integer arithmetic. */
|
|
|
|
|
|
|
|
#ifndef YYMAXDEPTH
|
|
|
|
# define YYMAXDEPTH 10000
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if YYERROR_VERBOSE
|
|
|
|
|
|
|
|
# ifndef yystrlen
|
|
|
|
# if defined __GLIBC__ && defined _STRING_H
|
|
|
|
# define yystrlen strlen
|
|
|
|
# else
|
|
|
|
/* Return the length of YYSTR. */
|
|
|
|
static YYSIZE_T
|
|
|
|
yystrlen (const char *yystr)
|
|
|
|
{
|
|
|
|
YYSIZE_T yylen;
|
|
|
|
for (yylen = 0; yystr[yylen]; yylen++)
|
|
|
|
continue;
|
|
|
|
return yylen;
|
|
|
|
}
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
|
|
|
|
# ifndef yystpcpy
|
|
|
|
# if defined __GLIBC__ && defined _STRING_H && defined _GNU_SOURCE
|
|
|
|
# define yystpcpy stpcpy
|
|
|
|
# else
|
|
|
|
/* Copy YYSRC to YYDEST, returning the address of the terminating '\0' in
|
|
|
|
YYDEST. */
|
|
|
|
static char *
|
|
|
|
yystpcpy (char *yydest, const char *yysrc)
|
|
|
|
{
|
|
|
|
char *yyd = yydest;
|
|
|
|
const char *yys = yysrc;
|
|
|
|
|
|
|
|
while ((*yyd++ = *yys++) != '\0')
|
|
|
|
continue;
|
|
|
|
|
|
|
|
return yyd - 1;
|
|
|
|
}
|
|
|
|
# endif
|
|
|
|
# endif
|
|
|
|
|
|
|
|
# ifndef yytnamerr
|
|
|
|
/* Copy to YYRES the contents of YYSTR after stripping away unnecessary
|
|
|
|
quotes and backslashes, so that it's suitable for yyerror. The
|
|
|
|
heuristic is that double-quoting is unnecessary unless the string
|
|
|
|
contains an apostrophe, a comma, or backslash (other than
|
|
|
|
backslash-backslash). YYSTR is taken from yytname. If YYRES is
|
|
|
|
null, do not copy; instead, return the length of what the result
|
|
|
|
would have been. */
|
|
|
|
static YYSIZE_T
|
|
|
|
yytnamerr (char *yyres, const char *yystr)
|
|
|
|
{
|
|
|
|
if (*yystr == '"')
|
|
|
|
{
|
|
|
|
YYSIZE_T yyn = 0;
|
|
|
|
char const *yyp = yystr;
|
|
|
|
|
|
|
|
for (;;)
|
|
|
|
switch (*++yyp)
|
|
|
|
{
|
|
|
|
case '\'':
|
|
|
|
case ',':
|
|
|
|
goto do_not_strip_quotes;
|
|
|
|
|
|
|
|
case '\\':
|
|
|
|
if (*++yyp != '\\')
|
|
|
|
goto do_not_strip_quotes;
|
|
|
|
/* Fall through. */
|
|
|
|
default:
|
|
|
|
if (yyres)
|
|
|
|
yyres[yyn] = *yyp;
|
|
|
|
yyn++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case '"':
|
|
|
|
if (yyres)
|
|
|
|
yyres[yyn] = '\0';
|
|
|
|
return yyn;
|
|
|
|
}
|
|
|
|
do_not_strip_quotes: ;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (! yyres)
|
|
|
|
return yystrlen (yystr);
|
|
|
|
|
|
|
|
return yystpcpy (yyres, yystr) - yyres;
|
|
|
|
}
|
|
|
|
# endif
|
|
|
|
|
|
|
|
/* Copy into *YYMSG, which is of size *YYMSG_ALLOC, an error message
|
|
|
|
about the unexpected token YYTOKEN for the state stack whose top is
|
|
|
|
YYSSP.
|
|
|
|
|
|
|
|
Return 0 if *YYMSG was successfully written. Return 1 if *YYMSG is
|
|
|
|
not large enough to hold the message. In that case, also set
|
|
|
|
*YYMSG_ALLOC to the required number of bytes. Return 2 if the
|
|
|
|
required number of bytes is too large to store. */
|
|
|
|
static int
|
|
|
|
yysyntax_error (YYSIZE_T *yymsg_alloc, char **yymsg,
|
|
|
|
yytype_int16 *yyssp, int yytoken)
|
|
|
|
{
|
|
|
|
YYSIZE_T yysize0 = yytnamerr (YY_NULLPTR, yytname[yytoken]);
|
|
|
|
YYSIZE_T yysize = yysize0;
|
|
|
|
enum { YYERROR_VERBOSE_ARGS_MAXIMUM = 5 };
|
|
|
|
/* Internationalized format string. */
|
|
|
|
const char *yyformat = YY_NULLPTR;
|
|
|
|
/* Arguments of yyformat. */
|
|
|
|
char const *yyarg[YYERROR_VERBOSE_ARGS_MAXIMUM];
|
|
|
|
/* Number of reported tokens (one for the "unexpected", one per
|
|
|
|
"expected"). */
|
|
|
|
int yycount = 0;
|
|
|
|
|
|
|
|
/* There are many possibilities here to consider:
|
|
|
|
- If this state is a consistent state with a default action, then
|
|
|
|
the only way this function was invoked is if the default action
|
|
|
|
is an error action. In that case, don't check for expected
|
|
|
|
tokens because there are none.
|
|
|
|
- The only way there can be no lookahead present (in yychar) is if
|
|
|
|
this state is a consistent state with a default action. Thus,
|
|
|
|
detecting the absence of a lookahead is sufficient to determine
|
|
|
|
that there is no unexpected or expected token to report. In that
|
|
|
|
case, just report a simple "syntax error".
|
|
|
|
- Don't assume there isn't a lookahead just because this state is a
|
|
|
|
consistent state with a default action. There might have been a
|
|
|
|
previous inconsistent state, consistent state with a non-default
|
|
|
|
action, or user semantic action that manipulated yychar.
|
|
|
|
- Of course, the expected token list depends on states to have
|
|
|
|
correct lookahead information, and it depends on the parser not
|
|
|
|
to perform extra reductions after fetching a lookahead from the
|
|
|
|
scanner and before detecting a syntax error. Thus, state merging
|
|
|
|
(from LALR or IELR) and default reductions corrupt the expected
|
|
|
|
token list. However, the list is correct for canonical LR with
|
|
|
|
one exception: it will still contain any token that will not be
|
|
|
|
accepted due to an error action in a later state.
|
|
|
|
*/
|
|
|
|
if (yytoken != YYEMPTY)
|
|
|
|
{
|
|
|
|
int yyn = yypact[*yyssp];
|
|
|
|
yyarg[yycount++] = yytname[yytoken];
|
|
|
|
if (!yypact_value_is_default (yyn))
|
|
|
|
{
|
|
|
|
/* Start YYX at -YYN if negative to avoid negative indexes in
|
|
|
|
YYCHECK. In other words, skip the first -YYN actions for
|
|
|
|
this state because they are default actions. */
|
|
|
|
int yyxbegin = yyn < 0 ? -yyn : 0;
|
|
|
|
/* Stay within bounds of both yycheck and yytname. */
|
|
|
|
int yychecklim = YYLAST - yyn + 1;
|
|
|
|
int yyxend = yychecklim < YYNTOKENS ? yychecklim : YYNTOKENS;
|
|
|
|
int yyx;
|
|
|
|
|
|
|
|
for (yyx = yyxbegin; yyx < yyxend; ++yyx)
|
|
|
|
if (yycheck[yyx + yyn] == yyx && yyx != YYTERROR
|
|
|
|
&& !yytable_value_is_error (yytable[yyx + yyn]))
|
|
|
|
{
|
|
|
|
if (yycount == YYERROR_VERBOSE_ARGS_MAXIMUM)
|
|
|
|
{
|
|
|
|
yycount = 1;
|
|
|
|
yysize = yysize0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
yyarg[yycount++] = yytname[yyx];
|
|
|
|
{
|
|
|
|
YYSIZE_T yysize1 = yysize + yytnamerr (YY_NULLPTR, yytname[yyx]);
|
|
|
|
if (! (yysize <= yysize1
|
|
|
|
&& yysize1 <= YYSTACK_ALLOC_MAXIMUM))
|
|
|
|
return 2;
|
|
|
|
yysize = yysize1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (yycount)
|
|
|
|
{
|
|
|
|
# define YYCASE_(N, S) \
|
|
|
|
case N: \
|
|
|
|
yyformat = S; \
|
|
|
|
break
|
|
|
|
YYCASE_(0, YY_("syntax error"));
|
|
|
|
YYCASE_(1, YY_("syntax error, unexpected %s"));
|
|
|
|
YYCASE_(2, YY_("syntax error, unexpected %s, expecting %s"));
|
|
|
|
YYCASE_(3, YY_("syntax error, unexpected %s, expecting %s or %s"));
|
|
|
|
YYCASE_(4, YY_("syntax error, unexpected %s, expecting %s or %s or %s"));
|
|
|
|
YYCASE_(5, YY_("syntax error, unexpected %s, expecting %s or %s or %s or %s"));
|
|
|
|
# undef YYCASE_
|
|
|
|
}
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
{
|
|
|
|
YYSIZE_T yysize1 = yysize + yystrlen (yyformat);
|
|
|
|
if (! (yysize <= yysize1 && yysize1 <= YYSTACK_ALLOC_MAXIMUM))
|
|
|
|
return 2;
|
|
|
|
yysize = yysize1;
|
|
|
|
}
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
if (*yymsg_alloc < yysize)
|
|
|
|
{
|
|
|
|
*yymsg_alloc = 2 * yysize;
|
|
|
|
if (! (yysize <= *yymsg_alloc
|
|
|
|
&& *yymsg_alloc <= YYSTACK_ALLOC_MAXIMUM))
|
|
|
|
*yymsg_alloc = YYSTACK_ALLOC_MAXIMUM;
|
|
|
|
return 1;
|
|
|
|
}
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Avoid sprintf, as that infringes on the user's name space.
|
|
|
|
Don't have undefined behavior even if the translation
|
|
|
|
produced a string with the wrong number of "%s"s. */
|
|
|
|
{
|
|
|
|
char *yyp = *yymsg;
|
|
|
|
int yyi = 0;
|
|
|
|
while ((*yyp = *yyformat) != '\0')
|
|
|
|
if (*yyp == '%' && yyformat[1] == 's' && yyi < yycount)
|
|
|
|
{
|
|
|
|
yyp += yytnamerr (yyp, yyarg[yyi++]);
|
|
|
|
yyformat += 2;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
yyp++;
|
|
|
|
yyformat++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif /* YYERROR_VERBOSE */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/*-----------------------------------------------.
|
|
|
|
| Release the memory associated to this symbol. |
|
|
|
|
`-----------------------------------------------*/
|
|
|
|
|
|
|
|
static void
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yydestruct (const char *yymsg, int yytype, YYSTYPE *yyvaluep)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYUSE (yyvaluep);
|
2020-07-30 07:59:19 +08:00
|
|
|
if (!yymsg)
|
|
|
|
yymsg = "Deleting";
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YY_SYMBOL_PRINT (yymsg, yytype, yyvaluep, yylocationp);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
YY_IGNORE_MAYBE_UNINITIALIZED_BEGIN
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYUSE (yytype);
|
2020-07-30 07:59:19 +08:00
|
|
|
YY_IGNORE_MAYBE_UNINITIALIZED_END
|
|
|
|
}
|
|
|
|
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
|
|
|
|
|
|
|
/* The lookahead symbol. */
|
2020-07-30 07:59:19 +08:00
|
|
|
int yychar;
|
|
|
|
|
|
|
|
/* The semantic value of the lookahead symbol. */
|
|
|
|
YYSTYPE yylval;
|
|
|
|
/* Number of syntax errors so far. */
|
|
|
|
int yynerrs;
|
|
|
|
|
|
|
|
|
|
|
|
/*----------.
|
|
|
|
| yyparse. |
|
|
|
|
`----------*/
|
|
|
|
|
|
|
|
int
|
|
|
|
yyparse (void)
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
int yystate;
|
2020-07-30 07:59:19 +08:00
|
|
|
/* Number of tokens to shift before error messages enabled. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
int yyerrstatus;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* The stacks and their tools:
|
|
|
|
'yyss': related to states.
|
|
|
|
'yyvs': related to semantic values.
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
Refer to the stacks through separate pointers, to allow yyoverflow
|
|
|
|
to reallocate them elsewhere. */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* The state stack. */
|
|
|
|
yytype_int16 yyssa[YYINITDEPTH];
|
|
|
|
yytype_int16 *yyss;
|
|
|
|
yytype_int16 *yyssp;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* The semantic value stack. */
|
2021-09-03 07:04:26 +08:00
|
|
|
YYSTYPE yyvsa[YYINITDEPTH];
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYSTYPE *yyvs;
|
|
|
|
YYSTYPE *yyvsp;
|
|
|
|
|
|
|
|
YYSIZE_T yystacksize;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
int yyn;
|
|
|
|
int yyresult;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Lookahead token as an internal (translated) token number. */
|
|
|
|
int yytoken = 0;
|
2020-07-30 07:59:19 +08:00
|
|
|
/* The variables used to return semantic value and location from the
|
|
|
|
action routines. */
|
|
|
|
YYSTYPE yyval;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if YYERROR_VERBOSE
|
|
|
|
/* Buffer for error messages, and its allocated size. */
|
|
|
|
char yymsgbuf[128];
|
|
|
|
char *yymsg = yymsgbuf;
|
|
|
|
YYSIZE_T yymsg_alloc = sizeof yymsgbuf;
|
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
#define YYPOPSTACK(N) (yyvsp -= (N), yyssp -= (N))
|
|
|
|
|
|
|
|
/* The number of symbols on the RHS of the reduced rule.
|
|
|
|
Keep to zero when no symbol should be popped. */
|
|
|
|
int yylen = 0;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yyssp = yyss = yyssa;
|
|
|
|
yyvsp = yyvs = yyvsa;
|
|
|
|
yystacksize = YYINITDEPTH;
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
YYDPRINTF ((stderr, "Starting parse\n"));
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yystate = 0;
|
|
|
|
yyerrstatus = 0;
|
|
|
|
yynerrs = 0;
|
2020-07-30 07:59:19 +08:00
|
|
|
yychar = YYEMPTY; /* Cause a token to be read. */
|
|
|
|
goto yysetstate;
|
|
|
|
|
|
|
|
/*------------------------------------------------------------.
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
| yynewstate -- Push a new state, which is found in yystate. |
|
2020-07-30 07:59:19 +08:00
|
|
|
`------------------------------------------------------------*/
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yynewstate:
|
2020-07-30 07:59:19 +08:00
|
|
|
/* In all cases, when you get here, the value and location stacks
|
|
|
|
have just been pushed. So pushing a state here evens the stacks. */
|
|
|
|
yyssp++;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yysetstate:
|
|
|
|
*yyssp = yystate;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
if (yyss + yystacksize - 1 <= yyssp)
|
|
|
|
{
|
|
|
|
/* Get the current used size of the three stacks, in elements. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYSIZE_T yysize = yyssp - yyss + 1;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#ifdef yyoverflow
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
/* Give user a chance to reallocate the stack. Use copies of
|
|
|
|
these so that the &'s don't force the real ones into
|
|
|
|
memory. */
|
|
|
|
YYSTYPE *yyvs1 = yyvs;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yytype_int16 *yyss1 = yyss;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* Each stack pointer address is followed by the size of the
|
|
|
|
data in use in that stack, in bytes. This used to be a
|
|
|
|
conditional around just the two extra args, but that might
|
|
|
|
be undefined if yyoverflow is a macro. */
|
|
|
|
yyoverflow (YY_("memory exhausted"),
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
&yyss1, yysize * sizeof (*yyssp),
|
|
|
|
&yyvs1, yysize * sizeof (*yyvsp),
|
2020-07-30 07:59:19 +08:00
|
|
|
&yystacksize);
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
yyss = yyss1;
|
|
|
|
yyvs = yyvs1;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#else /* no yyoverflow */
|
|
|
|
# ifndef YYSTACK_RELOCATE
|
|
|
|
goto yyexhaustedlab;
|
|
|
|
# else
|
2020-07-30 07:59:19 +08:00
|
|
|
/* Extend the stack our own way. */
|
|
|
|
if (YYMAXDEPTH <= yystacksize)
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
goto yyexhaustedlab;
|
2020-07-30 07:59:19 +08:00
|
|
|
yystacksize *= 2;
|
|
|
|
if (YYMAXDEPTH < yystacksize)
|
|
|
|
yystacksize = YYMAXDEPTH;
|
|
|
|
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yytype_int16 *yyss1 = yyss;
|
2020-07-30 07:59:19 +08:00
|
|
|
union yyalloc *yyptr =
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(union yyalloc *) YYSTACK_ALLOC (YYSTACK_BYTES (yystacksize));
|
2020-07-30 07:59:19 +08:00
|
|
|
if (! yyptr)
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
goto yyexhaustedlab;
|
2020-07-30 07:59:19 +08:00
|
|
|
YYSTACK_RELOCATE (yyss_alloc, yyss);
|
|
|
|
YYSTACK_RELOCATE (yyvs_alloc, yyvs);
|
|
|
|
# undef YYSTACK_RELOCATE
|
|
|
|
if (yyss1 != yyssa)
|
|
|
|
YYSTACK_FREE (yyss1);
|
|
|
|
}
|
|
|
|
# endif
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#endif /* no yyoverflow */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
yyssp = yyss + yysize - 1;
|
|
|
|
yyvsp = yyvs + yysize - 1;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYDPRINTF ((stderr, "Stack size increased to %lu\n",
|
|
|
|
(unsigned long int) yystacksize));
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
if (yyss + yystacksize - 1 <= yyssp)
|
|
|
|
YYABORT;
|
|
|
|
}
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYDPRINTF ((stderr, "Entering state %d\n", yystate));
|
2021-12-24 13:18:56 +08:00
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
if (yystate == YYFINAL)
|
|
|
|
YYACCEPT;
|
|
|
|
|
|
|
|
goto yybackup;
|
|
|
|
|
|
|
|
/*-----------.
|
|
|
|
| yybackup. |
|
|
|
|
`-----------*/
|
|
|
|
yybackup:
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
/* Do appropriate processing given the current state. Read a
|
|
|
|
lookahead token if we need one and don't already have one. */
|
|
|
|
|
|
|
|
/* First try to decide what to do without reference to lookahead token. */
|
|
|
|
yyn = yypact[yystate];
|
|
|
|
if (yypact_value_is_default (yyn))
|
|
|
|
goto yydefault;
|
|
|
|
|
|
|
|
/* Not known => get a lookahead token if don't already have one. */
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* YYCHAR is either YYEMPTY or YYEOF or a valid lookahead symbol. */
|
2020-07-30 07:59:19 +08:00
|
|
|
if (yychar == YYEMPTY)
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YYDPRINTF ((stderr, "Reading a token: "));
|
2020-07-30 07:59:19 +08:00
|
|
|
yychar = yylex ();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (yychar <= YYEOF)
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yychar = yytoken = YYEOF;
|
2020-07-30 07:59:19 +08:00
|
|
|
YYDPRINTF ((stderr, "Now at end of input.\n"));
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
yytoken = YYTRANSLATE (yychar);
|
|
|
|
YY_SYMBOL_PRINT ("Next token is", yytoken, &yylval, &yylloc);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* If the proper action on seeing token YYTOKEN is to reduce or to
|
|
|
|
detect an error, take that action. */
|
|
|
|
yyn += yytoken;
|
|
|
|
if (yyn < 0 || YYLAST < yyn || yycheck[yyn] != yytoken)
|
|
|
|
goto yydefault;
|
|
|
|
yyn = yytable[yyn];
|
|
|
|
if (yyn <= 0)
|
|
|
|
{
|
|
|
|
if (yytable_value_is_error (yyn))
|
|
|
|
goto yyerrlab;
|
|
|
|
yyn = -yyn;
|
|
|
|
goto yyreduce;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Count tokens shifted since error; after three, turn off error
|
|
|
|
status. */
|
|
|
|
if (yyerrstatus)
|
|
|
|
yyerrstatus--;
|
|
|
|
|
|
|
|
/* Shift the lookahead token. */
|
|
|
|
YY_SYMBOL_PRINT ("Shifting", yytoken, &yylval, &yylloc);
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
|
|
|
/* Discard the shifted token. */
|
|
|
|
yychar = YYEMPTY;
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
yystate = yyn;
|
|
|
|
YY_IGNORE_MAYBE_UNINITIALIZED_BEGIN
|
|
|
|
*++yyvsp = yylval;
|
|
|
|
YY_IGNORE_MAYBE_UNINITIALIZED_END
|
|
|
|
|
|
|
|
goto yynewstate;
|
|
|
|
|
|
|
|
|
|
|
|
/*-----------------------------------------------------------.
|
|
|
|
| yydefault -- do the default action for the current state. |
|
|
|
|
`-----------------------------------------------------------*/
|
|
|
|
yydefault:
|
|
|
|
yyn = yydefact[yystate];
|
|
|
|
if (yyn == 0)
|
|
|
|
goto yyerrlab;
|
|
|
|
goto yyreduce;
|
|
|
|
|
|
|
|
|
|
|
|
/*-----------------------------.
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
| yyreduce -- Do a reduction. |
|
2020-07-30 07:59:19 +08:00
|
|
|
`-----------------------------*/
|
|
|
|
yyreduce:
|
|
|
|
/* yyn is the number of a rule to reduce with. */
|
|
|
|
yylen = yyr2[yyn];
|
|
|
|
|
|
|
|
/* If YYLEN is nonzero, implement the default value of the action:
|
|
|
|
'$$ = $1'.
|
|
|
|
|
|
|
|
Otherwise, the following line sets YYVAL to garbage.
|
|
|
|
This behavior is undocumented and Bison
|
|
|
|
users should not rely upon it. Assigning to YYVAL
|
|
|
|
unconditionally makes the parser a bit smaller, and it avoids a
|
|
|
|
GCC warning that YYVAL may be used uninitialized. */
|
|
|
|
yyval = yyvsp[1-yylen];
|
|
|
|
|
|
|
|
|
|
|
|
YY_REDUCE_PRINT (yyn);
|
|
|
|
switch (yyn)
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 2:
|
|
|
|
#line 117 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ init_netcdf(); }
|
|
|
|
#line 1379 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 3:
|
|
|
|
#line 120 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if (derror_count == 0)
|
|
|
|
define_netcdf(netcdfname);
|
|
|
|
if (derror_count > 0)
|
|
|
|
exit(6);
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1390 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 4:
|
|
|
|
#line 128 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if (derror_count == 0)
|
|
|
|
close_netcdf();
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1399 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 11:
|
|
|
|
#line 143 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ if (int_val <= 0)
|
2020-07-30 07:59:19 +08:00
|
|
|
derror("dimension length must be positive");
|
|
|
|
dims[ndims].size = int_val;
|
|
|
|
ndims++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1409 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 12:
|
|
|
|
#line 149 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ /* for rare case where 2^31 < dimsize < 2^32 */
|
2020-07-30 07:59:19 +08:00
|
|
|
if (double_val <= 0)
|
|
|
|
derror("dimension length must be positive");
|
|
|
|
if (double_val > 4294967295.0)
|
|
|
|
derror("dimension too large");
|
|
|
|
if (double_val - (size_t) double_val > 0)
|
|
|
|
derror("dimension length must be an integer");
|
|
|
|
dims[ndims].size = (size_t) double_val;
|
|
|
|
ndims++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1424 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 13:
|
|
|
|
#line 160 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ if (rec_dim != -1)
|
2020-07-30 07:59:19 +08:00
|
|
|
derror("only one NC_UNLIMITED dimension allowed");
|
|
|
|
rec_dim = ndims; /* the unlimited (record) dimension */
|
|
|
|
dims[ndims].size = NC_UNLIMITED;
|
|
|
|
ndims++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1435 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 14:
|
|
|
|
#line 168 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
|
|
|
if ((yyvsp[0])->is_dim == 1) {
|
2020-07-30 07:59:19 +08:00
|
|
|
derror( "duplicate dimension declaration for %s",
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(yyvsp[0])->is_dim = 1;
|
|
|
|
(yyvsp[0])->dnum = ndims;
|
2020-07-30 07:59:19 +08:00
|
|
|
/* make sure dims array will hold dimensions */
|
|
|
|
grow_darray(ndims, /* must hold ndims+1 dims */
|
|
|
|
&dims); /* grow as needed */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
dims[ndims].name = (char *) emalloc(strlen((yyvsp[0])->name)+1);
|
|
|
|
(void) strcpy(dims[ndims].name, (yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
/* name for use in generated Fortran and C variables */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
dims[ndims].lname = decodify((yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1455 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 27:
|
|
|
|
#line 200 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ type_code = NC_BYTE; }
|
|
|
|
#line 1461 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 28:
|
|
|
|
#line 201 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ type_code = NC_CHAR; }
|
|
|
|
#line 1467 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 29:
|
|
|
|
#line 202 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ type_code = NC_SHORT; }
|
|
|
|
#line 1473 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 30:
|
|
|
|
#line 203 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ type_code = NC_INT; }
|
|
|
|
#line 1479 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 31:
|
|
|
|
#line 204 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ type_code = NC_FLOAT; }
|
|
|
|
#line 1485 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 32:
|
|
|
|
#line 205 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ type_code = NC_DOUBLE; }
|
|
|
|
#line 1491 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 35:
|
|
|
|
#line 211 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
static struct vars dummyvar;
|
|
|
|
|
|
|
|
dummyvar.name = "dummy";
|
|
|
|
dummyvar.type = NC_DOUBLE;
|
|
|
|
dummyvar.ndims = 0;
|
|
|
|
dummyvar.dims = 0;
|
|
|
|
dummyvar.fill_value.doublev = NC_FILL_DOUBLE;
|
|
|
|
dummyvar.has_data = 0;
|
|
|
|
|
|
|
|
nvdims = 0;
|
|
|
|
/* make sure variable not re-declared */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
if ((yyvsp[0])->is_var == 1) {
|
2020-07-30 07:59:19 +08:00
|
|
|
derror( "duplicate variable declaration for %s",
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(yyvsp[0])->is_var = 1;
|
|
|
|
(yyvsp[0])->vnum = nvars;
|
2020-07-30 07:59:19 +08:00
|
|
|
/* make sure vars array will hold variables */
|
|
|
|
grow_varray(nvars, /* must hold nvars+1 vars */
|
|
|
|
&vars); /* grow as needed */
|
|
|
|
vars[nvars] = dummyvar; /* to make Purify happy */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
vars[nvars].name = (char *) emalloc(strlen((yyvsp[0])->name)+1);
|
|
|
|
(void) strcpy(vars[nvars].name, (yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
/* name for use in generated Fortran and C variables */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
vars[nvars].lname = decodify((yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
vars[nvars].type = type_code;
|
|
|
|
/* set default fill value. You can override this with
|
|
|
|
* the variable attribute "_FillValue". */
|
|
|
|
nc_getfill(type_code, &vars[nvars].fill_value);
|
|
|
|
vars[nvars].has_data = 0; /* has no data (yet) */
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1528 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 36:
|
|
|
|
#line 244 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
vars[nvars].ndims = nvdims;
|
|
|
|
nvars++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1537 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 42:
|
|
|
|
#line 258 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if (nvdims >= NC_MAX_VAR_DIMS) {
|
|
|
|
derror("%s has too many dimensions",vars[nvars].name);
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
if ((yyvsp[0])->is_dim == 1)
|
|
|
|
dimnum = (yyvsp[0])->dnum;
|
2020-07-30 07:59:19 +08:00
|
|
|
else {
|
|
|
|
derror( "%s is not declared as a dimension",
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
dimnum = ndims;
|
|
|
|
}
|
|
|
|
if (rec_dim != -1 && dimnum == rec_dim && nvdims != 0) {
|
|
|
|
derror("unlimited dimension must be first");
|
|
|
|
}
|
|
|
|
grow_iarray(nvdims, /* must hold nvdims+1 ints */
|
|
|
|
&vars[nvars].dims); /* grow as needed */
|
|
|
|
vars[nvars].dims[nvdims] = dimnum;
|
|
|
|
nvdims++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1561 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 43:
|
|
|
|
#line 279 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
defatt();
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1569 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 44:
|
|
|
|
#line 283 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
equalatt();
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1577 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 45:
|
|
|
|
#line 288 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
defatt();
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1585 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 46:
|
|
|
|
#line 292 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
equalatt();
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1593 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 48:
|
|
|
|
#line 300 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
varnum = NC_GLOBAL; /* handle of "global" attribute */
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1601 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 49:
|
|
|
|
#line 306 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{ if ((yyvsp[0])->is_var == 1)
|
|
|
|
varnum = (yyvsp[0])->vnum;
|
2020-07-30 07:59:19 +08:00
|
|
|
else {
|
|
|
|
derror("%s not declared as a variable, fatal error",
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
(yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
YYABORT;
|
|
|
|
}
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1614 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 50:
|
|
|
|
#line 316 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
/* make sure atts array will hold attributes */
|
|
|
|
grow_aarray(natts, /* must hold natts+1 atts */
|
|
|
|
&atts); /* grow as needed */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
atts[natts].name = (char *) emalloc(strlen((yyvsp[0])->name)+1);
|
|
|
|
(void) strcpy(atts[natts].name,(yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
/* name for use in generated Fortran and C variables */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
atts[natts].lname = decodify((yyvsp[0])->name);
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1628 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 53:
|
|
|
|
#line 330 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if (valtype == NC_UNSPECIFIED)
|
|
|
|
valtype = atype_code;
|
|
|
|
if (valtype != atype_code)
|
|
|
|
derror("values for attribute must be all of same type");
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1639 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 54:
|
|
|
|
#line 339 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_CHAR;
|
|
|
|
*char_valp++ = char_val;
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1649 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 55:
|
|
|
|
#line 345 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_CHAR;
|
|
|
|
{
|
|
|
|
/* don't null-terminate attribute strings */
|
|
|
|
size_t len = strlen(termstring);
|
|
|
|
if (len == 0) /* need null if that's only value */
|
|
|
|
len = 1;
|
|
|
|
(void)strncpy(char_valp,termstring,len);
|
|
|
|
valnum += len;
|
|
|
|
char_valp += len;
|
|
|
|
}
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1666 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 56:
|
|
|
|
#line 358 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_BYTE;
|
|
|
|
*byte_valp++ = byte_val;
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1676 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 57:
|
|
|
|
#line 364 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_SHORT;
|
|
|
|
*short_valp++ = short_val;
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1686 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 58:
|
|
|
|
#line 370 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_INT;
|
|
|
|
*int_valp++ = int_val;
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1696 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 59:
|
|
|
|
#line 376 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_FLOAT;
|
|
|
|
*float_valp++ = float_val;
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1706 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 60:
|
|
|
|
#line 382 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_DOUBLE;
|
|
|
|
*double_valp++ = double_val;
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1716 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 66:
|
|
|
|
#line 398 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
valtype = vars[varnum].type; /* variable type */
|
|
|
|
valnum = 0; /* values accumulated for variable */
|
|
|
|
vars[varnum].has_data = 1;
|
|
|
|
/* compute dimensions product */
|
|
|
|
var_size = nctypesize(valtype);
|
|
|
|
if (vars[varnum].ndims == 0) { /* scalar */
|
|
|
|
var_len = 1;
|
|
|
|
} else if (vars[varnum].dims[0] == rec_dim) {
|
|
|
|
var_len = 1; /* one record for unlimited vars */
|
|
|
|
} else {
|
|
|
|
var_len = dims[vars[varnum].dims[0]].size;
|
|
|
|
}
|
|
|
|
for(dimnum = 1; dimnum < vars[varnum].ndims; dimnum++)
|
|
|
|
var_len = var_len*dims[vars[varnum].dims[dimnum]].size;
|
|
|
|
/* allocate memory for variable data */
|
|
|
|
if (var_len*var_size != (size_t)(var_len*var_size)) {
|
|
|
|
derror("variable %s too large for memory",
|
|
|
|
vars[varnum].name);
|
|
|
|
exit(9);
|
|
|
|
}
|
|
|
|
rec_len = var_len;
|
|
|
|
rec_start = malloc ((size_t)(rec_len*var_size));
|
|
|
|
if (rec_start == 0) {
|
|
|
|
derror ("out of memory\n");
|
|
|
|
exit(3);
|
|
|
|
}
|
|
|
|
rec_cur = rec_start;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
char_valp = (char *) rec_start;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
byte_valp = (signed char *) rec_start;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
short_valp = (short *) rec_start;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
int_valp = (int *) rec_start;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
float_valp = (float *) rec_start;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
double_valp = (double *) rec_start;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1771 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 67:
|
|
|
|
#line 449 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if (valnum < var_len) { /* leftovers */
|
|
|
|
nc_fill(valtype,
|
|
|
|
var_len - valnum,
|
|
|
|
rec_cur,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
}
|
|
|
|
/* put out var_len values */
|
|
|
|
/* vars[varnum].nrecs = valnum / rec_len; */
|
|
|
|
vars[varnum].nrecs = var_len / rec_len;
|
|
|
|
if (derror_count == 0)
|
|
|
|
put_variable(rec_start);
|
|
|
|
free ((char *) rec_start);
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1790 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 70:
|
|
|
|
#line 468 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if(valnum >= var_len) {
|
|
|
|
if (vars[varnum].dims[0] != rec_dim) { /* not recvar */
|
|
|
|
derror("too many values for this variable, %d >= %d",
|
|
|
|
valnum, var_len);
|
|
|
|
exit (4);
|
|
|
|
} else { /* a record variable, so grow data
|
|
|
|
container and increment var_len by
|
|
|
|
multiple of record size */
|
|
|
|
ptrdiff_t rec_inc = (char *)rec_cur
|
|
|
|
- (char *)rec_start;
|
|
|
|
var_len = rec_len * (1 + valnum / rec_len);
|
|
|
|
rec_start = erealloc(rec_start, var_len*var_size);
|
|
|
|
rec_cur = (char *)rec_start + rec_inc;
|
|
|
|
char_valp = (char *) rec_cur;
|
|
|
|
byte_valp = (signed char *) rec_cur;
|
|
|
|
short_valp = (short *) rec_cur;
|
|
|
|
int_valp = (int *) rec_cur;
|
|
|
|
float_valp = (float *) rec_cur;
|
|
|
|
double_valp = (double *) rec_cur;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
not_a_string = 1;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1819 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 71:
|
|
|
|
#line 493 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
if (not_a_string) {
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
rec_cur = (void *) char_valp;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
rec_cur = (void *) byte_valp;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
rec_cur = (void *) short_valp;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
rec_cur = (void *) int_valp;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
rec_cur = (void *) float_valp;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
rec_cur = (void *) double_valp;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1849 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 72:
|
|
|
|
#line 521 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_CHAR;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
*char_valp++ = char_val;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
*byte_valp++ = char_val;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
*short_valp++ = char_val;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
*int_valp++ = char_val;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
*float_valp++ = char_val;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
*double_valp++ = char_val;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1879 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 73:
|
|
|
|
#line 547 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
not_a_string = 0;
|
|
|
|
atype_code = NC_CHAR;
|
|
|
|
{
|
|
|
|
size_t len = strlen(termstring);
|
|
|
|
|
|
|
|
if(valnum + len > var_len) {
|
|
|
|
if (vars[varnum].dims[0] != rec_dim) {
|
|
|
|
derror("too many values for this variable, %d>%d",
|
|
|
|
valnum+len, var_len);
|
|
|
|
exit (5);
|
|
|
|
} else {/* a record variable so grow it */
|
|
|
|
ptrdiff_t rec_inc = (char *)rec_cur
|
|
|
|
- (char *)rec_start;
|
|
|
|
var_len += rec_len * (len + valnum - var_len)/rec_len;
|
|
|
|
rec_start = erealloc(rec_start, var_len*var_size);
|
|
|
|
rec_cur = (char *)rec_start + rec_inc;
|
|
|
|
char_valp = (char *) rec_cur;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
{
|
|
|
|
int ld;
|
|
|
|
size_t i, sl;
|
|
|
|
(void)strncpy(char_valp,termstring,len);
|
|
|
|
ld = vars[varnum].ndims-1;
|
|
|
|
if (ld > 0) {/* null-fill to size of last dim */
|
|
|
|
sl = dims[vars[varnum].dims[ld]].size;
|
|
|
|
for (i =len;i<sl;i++)
|
|
|
|
char_valp[i] = '\0';
|
|
|
|
if (sl < len)
|
|
|
|
sl = len;
|
|
|
|
valnum += sl;
|
|
|
|
char_valp += sl;
|
|
|
|
} else { /* scalar or 1D strings */
|
|
|
|
valnum += len;
|
|
|
|
char_valp += len;
|
|
|
|
}
|
|
|
|
rec_cur = (void *) char_valp;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
case NC_SHORT:
|
|
|
|
case NC_INT:
|
|
|
|
case NC_FLOAT:
|
|
|
|
case NC_DOUBLE:
|
|
|
|
derror("string value invalid for %s variable",
|
|
|
|
nctype(valtype));
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1938 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 74:
|
|
|
|
#line 602 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_BYTE;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
*char_valp++ = byte_val;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
*byte_valp++ = byte_val;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
*short_valp++ = byte_val;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
*int_valp++ = byte_val;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
*float_valp++ = byte_val;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
*double_valp++ = byte_val;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1968 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 75:
|
|
|
|
#line 628 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_SHORT;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
*char_valp++ = short_val;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
*byte_valp++ = short_val;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
*short_valp++ = short_val;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
*int_valp++ = short_val;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
*float_valp++ = short_val;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
*double_valp++ = short_val;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 1998 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 76:
|
|
|
|
#line 654 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_INT;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
*char_valp++ = int_val;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
*byte_valp++ = int_val;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
*short_valp++ = int_val;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
*int_valp++ = int_val;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
*float_valp++ = int_val;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
*double_valp++ = int_val;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 2028 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 77:
|
|
|
|
#line 680 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_FLOAT;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
*char_valp++ = float_val;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
*byte_valp++ = float_val;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
*short_valp++ = float_val;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
*int_valp++ = float_val;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
*float_valp++ = float_val;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
*double_valp++ = float_val;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 2058 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 78:
|
|
|
|
#line 706 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
atype_code = NC_DOUBLE;
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
*char_valp++ = double_val;
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
*byte_valp++ = double_val;
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
*short_valp++ = double_val;
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
*int_valp++ = double_val;
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
if (double_val == NC_FILL_DOUBLE)
|
|
|
|
*float_valp++ = NC_FILL_FLOAT;
|
|
|
|
else
|
|
|
|
*float_valp++ = double_val;
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
*double_valp++ = double_val;
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 2091 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
case 79:
|
|
|
|
#line 735 "ncgen.y" /* yacc.c:1646 */
|
|
|
|
{
|
2020-07-30 07:59:19 +08:00
|
|
|
/* store fill_value */
|
|
|
|
switch (valtype) {
|
|
|
|
case NC_CHAR:
|
|
|
|
nc_fill(valtype, 1, (void *)char_valp++,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
break;
|
|
|
|
case NC_BYTE:
|
|
|
|
nc_fill(valtype, 1, (void *)byte_valp++,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
break;
|
|
|
|
case NC_SHORT:
|
|
|
|
nc_fill(valtype, 1, (void *)short_valp++,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
break;
|
|
|
|
case NC_INT:
|
|
|
|
nc_fill(valtype, 1, (void *)int_valp++,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
break;
|
|
|
|
case NC_FLOAT:
|
|
|
|
nc_fill(valtype, 1, (void *)float_valp++,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
break;
|
|
|
|
case NC_DOUBLE:
|
|
|
|
nc_fill(valtype, 1, (void *)double_valp++,
|
|
|
|
vars[varnum].fill_value);
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
valnum++;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 2127 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 2131 "ncgeny.c" /* yacc.c:1646 */
|
2020-07-30 07:59:19 +08:00
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
/* User semantic actions sometimes alter yychar, and that requires
|
|
|
|
that yytoken be updated with the new translation. We take the
|
|
|
|
approach of translating immediately before every use of yytoken.
|
|
|
|
One alternative is translating here after every semantic action,
|
|
|
|
but that translation would be missed if the semantic action invokes
|
|
|
|
YYABORT, YYACCEPT, or YYERROR immediately after altering yychar or
|
|
|
|
if it invokes YYBACKUP. In the case of YYABORT or YYACCEPT, an
|
|
|
|
incorrect destructor might then be invoked immediately. In the
|
|
|
|
case of YYERROR or YYBACKUP, subsequent parser actions might lead
|
|
|
|
to an incorrect destructor call or verbose syntax error message
|
|
|
|
before the lookahead is translated. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YY_SYMBOL_PRINT ("-> $$ =", yyr1[yyn], &yyval, &yyloc);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
YYPOPSTACK (yylen);
|
|
|
|
yylen = 0;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YY_STACK_PRINT (yyss, yyssp);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
*++yyvsp = yyval;
|
|
|
|
|
|
|
|
/* Now 'shift' the result of the reduction. Determine what state
|
|
|
|
that goes to, based on the state we popped back to and the rule
|
|
|
|
number reduced by. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
|
|
|
yyn = yyr1[yyn];
|
|
|
|
|
|
|
|
yystate = yypgoto[yyn - YYNTOKENS] + *yyssp;
|
|
|
|
if (0 <= yystate && yystate <= YYLAST && yycheck[yystate] == *yyssp)
|
|
|
|
yystate = yytable[yystate];
|
|
|
|
else
|
|
|
|
yystate = yydefgoto[yyn - YYNTOKENS];
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
goto yynewstate;
|
|
|
|
|
|
|
|
|
|
|
|
/*--------------------------------------.
|
|
|
|
| yyerrlab -- here on detecting error. |
|
|
|
|
`--------------------------------------*/
|
|
|
|
yyerrlab:
|
|
|
|
/* Make sure we have latest lookahead translation. See comments at
|
|
|
|
user semantic actions for why this is necessary. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yytoken = yychar == YYEMPTY ? YYEMPTY : YYTRANSLATE (yychar);
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
/* If not already recovering from an error, report this error. */
|
|
|
|
if (!yyerrstatus)
|
|
|
|
{
|
|
|
|
++yynerrs;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if ! YYERROR_VERBOSE
|
2020-07-30 07:59:19 +08:00
|
|
|
yyerror (YY_("syntax error"));
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#else
|
|
|
|
# define YYSYNTAX_ERROR yysyntax_error (&yymsg_alloc, &yymsg, \
|
|
|
|
yyssp, yytoken)
|
|
|
|
{
|
|
|
|
char const *yymsgp = YY_("syntax error");
|
|
|
|
int yysyntax_error_status;
|
|
|
|
yysyntax_error_status = YYSYNTAX_ERROR;
|
|
|
|
if (yysyntax_error_status == 0)
|
|
|
|
yymsgp = yymsg;
|
|
|
|
else if (yysyntax_error_status == 1)
|
|
|
|
{
|
|
|
|
if (yymsg != yymsgbuf)
|
|
|
|
YYSTACK_FREE (yymsg);
|
|
|
|
yymsg = (char *) YYSTACK_ALLOC (yymsg_alloc);
|
|
|
|
if (!yymsg)
|
|
|
|
{
|
|
|
|
yymsg = yymsgbuf;
|
|
|
|
yymsg_alloc = sizeof yymsgbuf;
|
|
|
|
yysyntax_error_status = 2;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
yysyntax_error_status = YYSYNTAX_ERROR;
|
|
|
|
yymsgp = yymsg;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
yyerror (yymsgp);
|
|
|
|
if (yysyntax_error_status == 2)
|
|
|
|
goto yyexhaustedlab;
|
|
|
|
}
|
|
|
|
# undef YYSYNTAX_ERROR
|
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
}
|
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
if (yyerrstatus == 3)
|
|
|
|
{
|
|
|
|
/* If just tried and failed to reuse lookahead token after an
|
|
|
|
error, discard it. */
|
|
|
|
|
|
|
|
if (yychar <= YYEOF)
|
|
|
|
{
|
|
|
|
/* Return failure if at end of input. */
|
|
|
|
if (yychar == YYEOF)
|
|
|
|
YYABORT;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
yydestruct ("Error: discarding",
|
|
|
|
yytoken, &yylval);
|
|
|
|
yychar = YYEMPTY;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Else will try to reuse lookahead token after shifting the error
|
|
|
|
token. */
|
|
|
|
goto yyerrlab1;
|
|
|
|
|
|
|
|
|
|
|
|
/*---------------------------------------------------.
|
|
|
|
| yyerrorlab -- error raised explicitly by YYERROR. |
|
|
|
|
`---------------------------------------------------*/
|
|
|
|
yyerrorlab:
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
|
|
|
|
/* Pacify compilers like GCC when the user code never invokes
|
|
|
|
YYERROR and the label yyerrorlab therefore never appears in user
|
|
|
|
code. */
|
|
|
|
if (/*CONSTCOND*/ 0)
|
|
|
|
goto yyerrorlab;
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
/* Do not reclaim the symbols of the rule whose action triggered
|
|
|
|
this YYERROR. */
|
|
|
|
YYPOPSTACK (yylen);
|
|
|
|
yylen = 0;
|
|
|
|
YY_STACK_PRINT (yyss, yyssp);
|
|
|
|
yystate = *yyssp;
|
|
|
|
goto yyerrlab1;
|
|
|
|
|
|
|
|
|
|
|
|
/*-------------------------------------------------------------.
|
|
|
|
| yyerrlab1 -- common code for both syntax error and YYERROR. |
|
|
|
|
`-------------------------------------------------------------*/
|
|
|
|
yyerrlab1:
|
|
|
|
yyerrstatus = 3; /* Each real token shifted decrements this. */
|
|
|
|
|
|
|
|
for (;;)
|
|
|
|
{
|
|
|
|
yyn = yypact[yystate];
|
|
|
|
if (!yypact_value_is_default (yyn))
|
|
|
|
{
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yyn += YYTERROR;
|
|
|
|
if (0 <= yyn && yyn <= YYLAST && yycheck[yyn] == YYTERROR)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
yyn = yytable[yyn];
|
|
|
|
if (0 < yyn)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Pop the current state because it cannot handle the error token. */
|
|
|
|
if (yyssp == yyss)
|
|
|
|
YYABORT;
|
|
|
|
|
|
|
|
|
|
|
|
yydestruct ("Error: popping",
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yystos[yystate], yyvsp);
|
2020-07-30 07:59:19 +08:00
|
|
|
YYPOPSTACK (1);
|
|
|
|
yystate = *yyssp;
|
|
|
|
YY_STACK_PRINT (yyss, yyssp);
|
|
|
|
}
|
|
|
|
|
|
|
|
YY_IGNORE_MAYBE_UNINITIALIZED_BEGIN
|
|
|
|
*++yyvsp = yylval;
|
|
|
|
YY_IGNORE_MAYBE_UNINITIALIZED_END
|
|
|
|
|
|
|
|
|
|
|
|
/* Shift the error token. */
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
YY_SYMBOL_PRINT ("Shifting", yystos[yyn], yyvsp, yylsp);
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
yystate = yyn;
|
|
|
|
goto yynewstate;
|
|
|
|
|
|
|
|
|
|
|
|
/*-------------------------------------.
|
|
|
|
| yyacceptlab -- YYACCEPT comes here. |
|
|
|
|
`-------------------------------------*/
|
|
|
|
yyacceptlab:
|
|
|
|
yyresult = 0;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
goto yyreturn;
|
2021-09-03 07:04:26 +08:00
|
|
|
|
2020-07-30 07:59:19 +08:00
|
|
|
/*-----------------------------------.
|
|
|
|
| yyabortlab -- YYABORT comes here. |
|
|
|
|
`-----------------------------------*/
|
|
|
|
yyabortlab:
|
|
|
|
yyresult = 1;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
goto yyreturn;
|
2021-09-03 07:04:26 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if !defined yyoverflow || YYERROR_VERBOSE
|
|
|
|
/*-------------------------------------------------.
|
|
|
|
| yyexhaustedlab -- memory exhaustion comes here. |
|
|
|
|
`-------------------------------------------------*/
|
2020-07-30 07:59:19 +08:00
|
|
|
yyexhaustedlab:
|
|
|
|
yyerror (YY_("memory exhausted"));
|
|
|
|
yyresult = 2;
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
/* Fall through. */
|
|
|
|
#endif
|
2021-09-03 07:04:26 +08:00
|
|
|
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yyreturn:
|
2020-07-30 07:59:19 +08:00
|
|
|
if (yychar != YYEMPTY)
|
|
|
|
{
|
|
|
|
/* Make sure we have latest lookahead translation. See comments at
|
|
|
|
user semantic actions for why this is necessary. */
|
|
|
|
yytoken = YYTRANSLATE (yychar);
|
|
|
|
yydestruct ("Cleanup: discarding lookahead",
|
|
|
|
yytoken, &yylval);
|
|
|
|
}
|
|
|
|
/* Do not reclaim the symbols of the rule whose action triggered
|
|
|
|
this YYABORT or YYACCEPT. */
|
|
|
|
YYPOPSTACK (yylen);
|
|
|
|
YY_STACK_PRINT (yyss, yyssp);
|
|
|
|
while (yyssp != yyss)
|
|
|
|
{
|
|
|
|
yydestruct ("Cleanup: popping",
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
yystos[*yyssp], yyvsp);
|
2020-07-30 07:59:19 +08:00
|
|
|
YYPOPSTACK (1);
|
|
|
|
}
|
|
|
|
#ifndef yyoverflow
|
|
|
|
if (yyss != yyssa)
|
|
|
|
YYSTACK_FREE (yyss);
|
|
|
|
#endif
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#if YYERROR_VERBOSE
|
|
|
|
if (yymsg != yymsgbuf)
|
|
|
|
YYSTACK_FREE (yymsg);
|
|
|
|
#endif
|
2020-07-30 07:59:19 +08:00
|
|
|
return yyresult;
|
|
|
|
}
|
Fix various problem around VLEN's
re: https://github.com/Unidata/netcdf-c/issues/541
re: https://github.com/Unidata/netcdf-c/issues/1208
re: https://github.com/Unidata/netcdf-c/issues/2078
re: https://github.com/Unidata/netcdf-c/issues/2041
re: https://github.com/Unidata/netcdf-c/issues/2143
For a long time, there have been known problems with the
management of complex types containing VLENs. This also
involves the string type because it is stored as a VLEN of
chars.
This PR (mostly) fixes this problem. But note that it adds new
functions to netcdf.h (see below) and this may require bumping
the .so number. These new functions can be removed, if desired,
in favor of functions in netcdf_aux.h, but netcdf.h seems the
better place for them because they are intended as alternatives
to the nc_free_vlen and nc_free_string functions already in
netcdf.h.
The term complex type refers to any type that directly or
transitively references a VLEN type. So an array of VLENS, a
compound with a VLEN field, and so on.
In order to properly handle instances of these complex types, it
is necessary to have function that can recursively walk
instances of such types to perform various actions on them. The
term "deep" is also used to mean recursive.
At the moment, the two operations needed by the netcdf library are:
* free'ing an instance of the complex type
* copying an instance of the complex type.
The current library does only shallow free and shallow copy of
complex types. This means that only the top level is properly
free'd or copied, but deep internal blocks in the instance are
not touched.
Note that the term "vector" will be used to mean a contiguous (in
memory) sequence of instances of some type. Given an array with,
say, dimensions 2 X 3 X 4, this will be stored in memory as a
vector of length 2*3*4=24 instances.
The use cases are primarily these.
## nc_get_vars
Suppose one is reading a vector of instances using nc_get_vars
(or nc_get_vara or nc_get_var, etc.). These functions will
return the vector in the top-level memory provided. All
interior blocks (form nested VLEN or strings) will have been
dynamically allocated.
After using this vector of instances, it is necessary to free
(aka reclaim) the dynamically allocated memory, otherwise a
memory leak occurs. So, the recursive reclaim function is used
to walk the returned instance vector and do a deep reclaim of
the data.
Currently functions are defined in netcdf.h that are supposed to
handle this: nc_free_vlen(), nc_free_vlens(), and
nc_free_string(). Unfortunately, these functions only do a
shallow free, so deeply nested instances are not properly
handled by them.
Note that internally, the provided data is immediately written so
there is no need to copy it. But the caller may need to reclaim the
data it passed into the function.
## nc_put_att
Suppose one is writing a vector of instances as the data of an attribute
using, say, nc_put_att.
Internally, the incoming attribute data must be copied and stored
so that changes/reclamation of the input data will not affect
the attribute.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. As a result, one sees effects such as described
in Github Issue https://github.com/Unidata/netcdf-c/issues/2143.
Also, after defining the attribute, it may be necessary for the user
to free the data that was provided as input to nc_put_att().
## nc_get_att
Suppose one is reading a vector of instances as the data of an attribute
using, say, nc_get_att.
Internally, the existing attribute data must be copied and returned
to the caller, and the caller is responsible for reclaiming
the returned data.
Again, the code inside the netcdf library does only shallow copying
rather than deep copy. So this can lead to memory leaks and errors
because the deep data is shared between the library and the user.
# Solution
The solution is to build properly recursive reclaim and copy
functions and use those as needed.
These recursive functions are defined in libdispatch/dinstance.c
and their signatures are defined in include/netcdf.h.
For back compatibility, corresponding "ncaux_XXX" functions
are defined in include/netcdf_aux.h.
````
int nc_reclaim_data(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_reclaim_data_all(int ncid, nc_type xtypeid, void* memory, size_t count);
int nc_copy_data(int ncid, nc_type xtypeid, const void* memory, size_t count, void* copy);
int nc_copy_data_all(int ncid, nc_type xtypeid, const void* memory, size_t count, void** copyp);
````
There are two variants. The first two, nc_reclaim_data() and
nc_copy_data(), assume the top-level vector is managed by the
caller. For reclaim, this is so the user can use, for example, a
statically allocated vector. For copy, it assumes the user
provides the space into which the copy is stored.
The second two, nc_reclaim_data_all() and
nc_copy_data_all(), allows the functions to manage the
top-level. So for nc_reclaim_data_all, the top level is
assumed to be dynamically allocated and will be free'd by
nc_reclaim_data_all(). The nc_copy_data_all() function
will allocate the top level and return a pointer to it to the
user. The user can later pass that pointer to
nc_reclaim_data_all() to reclaim the instance(s).
# Internal Changes
The netcdf-c library internals are changed to use the proper
reclaim and copy functions. It turns out that the places where
these functions are needed is quite pervasive in the netcdf-c
library code. Using these functions also allows some
simplification of the code since the stdata and vldata fields of
NC_ATT_INFO are no longer needed. Currently this is commented
out using the SEPDATA \#define macro. When any bugs are largely
fixed, all this code will be removed.
# Known Bugs
1. There is still one known failure that has not been solved.
All the failures revolve around some variant of this .cdl file.
The proximate cause of failure is the use of a VLEN FillValue.
````
netcdf x {
types:
float(*) row_of_floats ;
dimensions:
m = 5 ;
variables:
row_of_floats ragged_array(m) ;
row_of_floats ragged_array:_FillValue = {-999} ;
data:
ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
{40, 41}, _ ;
}
````
When a solution is found, I will either add it to this PR or post a new PR.
# Related Changes
* Mark nc_free_vlen(s) as deprecated in favor of ncaux_reclaim_data.
* Remove the --enable-unfixed-memory-leaks option.
* Remove the NC_VLENS_NOTEST code that suppresses some vlen tests.
* Document this change in docs/internal.md
* Disable the tst_vlen_data test in ncdump/tst_nccopy4.sh.
* Mark types as fixed size or not (transitively) to optimize the reclaim
and copy functions.
# Misc. Changes
* Make Doxygen process libdispatch/daux.c
* Make sure the NC_ATT_INFO_T.container field is set.
2022-01-09 09:30:00 +08:00
|
|
|
#line 770 "ncgen.y" /* yacc.c:1906 */
|
2020-07-30 07:59:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
/* HELPER PROGRAMS */
|
2022-03-02 12:21:24 +08:00
|
|
|
void defatt(void)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
valnum = 0;
|
|
|
|
valtype = NC_UNSPECIFIED;
|
|
|
|
/* get a large block for attributes, realloc later */
|
|
|
|
att_space = emalloc(MAX_NC_ATTSIZE);
|
|
|
|
/* make all kinds of pointers point to it */
|
|
|
|
char_valp = (char *) att_space;
|
|
|
|
byte_valp = (signed char *) att_space;
|
|
|
|
short_valp = (short *) att_space;
|
|
|
|
int_valp = (int *) att_space;
|
|
|
|
float_valp = (float *) att_space;
|
|
|
|
double_valp = (double *) att_space;
|
|
|
|
}
|
|
|
|
|
2022-03-02 12:21:24 +08:00
|
|
|
void equalatt(void)
|
2020-07-30 07:59:19 +08:00
|
|
|
{
|
|
|
|
/* check if duplicate attribute for this var */
|
|
|
|
int i;
|
|
|
|
for(i=0; i<natts; i++) { /* expensive */
|
|
|
|
if(atts[i].var == varnum &&
|
|
|
|
STREQ(atts[i].name,atts[natts].name)) {
|
|
|
|
derror("duplicate attribute %s:%s",
|
|
|
|
vars[varnum].name,atts[natts].name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
atts[natts].var = varnum ;
|
|
|
|
atts[natts].type = valtype;
|
|
|
|
atts[natts].len = valnum;
|
|
|
|
/* shrink space down to what was really needed */
|
|
|
|
att_space = erealloc(att_space, valnum*nctypesize(valtype));
|
|
|
|
atts[natts].val = att_space;
|
|
|
|
if (STREQ(atts[natts].name, _FillValue) &&
|
|
|
|
atts[natts].var != NC_GLOBAL) {
|
|
|
|
nc_putfill(atts[natts].type,atts[natts].val,
|
|
|
|
&vars[atts[natts].var].fill_value);
|
|
|
|
if(atts[natts].type != vars[atts[natts].var].type) {
|
|
|
|
derror("variable %s: %s type mismatch",
|
|
|
|
vars[atts[natts].var].name, _FillValue);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
natts++;
|
|
|
|
}
|
|
|
|
/* PROGRAMS */
|
|
|
|
|
|
|
|
#ifdef vms
|
|
|
|
void
|
|
|
|
#else
|
|
|
|
int
|
|
|
|
#endif
|
|
|
|
yyerror( /* called for yacc syntax error */
|
|
|
|
char *s)
|
|
|
|
{
|
|
|
|
derror(s);
|
|
|
|
#ifndef vms
|
|
|
|
return -1;
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
/* undefine yywrap macro, in case we are using bison instead of yacc */
|
|
|
|
#ifdef yywrap
|
|
|
|
#undef yywrap
|
|
|
|
#endif
|
|
|
|
|
|
|
|
int
|
|
|
|
ncgwrap(void) /* returns 1 on EOF if no more input */
|
|
|
|
{
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* Symbol table operations for ncgen tool */
|
|
|
|
|
|
|
|
/* Find CDL name in symbol table (linear search). Note, this has a
|
|
|
|
* side-effect: it handles escape characters in the name, deleting
|
|
|
|
* single escape characters from the CDL name, before looking it up.
|
|
|
|
*/
|
|
|
|
YYSTYPE lookup(char *sname)
|
|
|
|
{
|
|
|
|
YYSTYPE sp;
|
|
|
|
deescapify(sname); /* delete escape chars from names,
|
|
|
|
* e.g. 'ab\:cd\ ef' becomes
|
|
|
|
* 'ab:cd ef' */
|
|
|
|
for (sp = symlist; sp != (YYSTYPE) 0; sp = sp -> next)
|
|
|
|
if (STREQ(sp -> name, sname)) {
|
|
|
|
return sp;
|
|
|
|
}
|
|
|
|
return 0; /* 0 ==> not found */
|
|
|
|
}
|
|
|
|
|
|
|
|
YYSTYPE install( /* install sname in symbol table */
|
|
|
|
const char *sname)
|
|
|
|
{
|
|
|
|
YYSTYPE sp;
|
|
|
|
|
|
|
|
sp = (YYSTYPE) emalloc (sizeof (struct Symbol));
|
|
|
|
sp -> name = (char *) emalloc (strlen (sname) + 1);/* +1 for '\0' */
|
|
|
|
(void) strcpy (sp -> name, sname);
|
|
|
|
sp -> next = symlist; /* put at front of list */
|
|
|
|
sp -> is_dim = 0;
|
|
|
|
sp -> is_var = 0;
|
|
|
|
sp -> is_att = 0;
|
|
|
|
symlist = sp;
|
|
|
|
return sp;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
clearout(void) /* reset symbol table to empty */
|
|
|
|
{
|
|
|
|
YYSTYPE sp, tp;
|
|
|
|
for (sp = symlist; sp != (YYSTYPE) 0;) {
|
|
|
|
tp = sp -> next;
|
|
|
|
free (sp -> name);
|
|
|
|
free ((char *) sp);
|
|
|
|
sp = tp;
|
|
|
|
}
|
|
|
|
symlist = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* get lexical input routine generated by lex */
|
|
|
|
|
|
|
|
/* Keep compile quiet */
|
|
|
|
#define YY_NO_UNPUT
|
|
|
|
#define YY_NO_INPUT
|
|
|
|
|
|
|
|
#include "ncgenl.c"
|